Knowledge and Justification 9781400870738

Table of contents :
Cover
Contents
Preface
1. What Is an Epistemological Problem?
2. The Structure of Epistemic Justification
3. Theories of Perceptual Knowledge
4. Incorrigibility
5. Perceptual Attributes
6. The Reidentification of Physical Things
7. Memory and Historical Knowledge
8. Induction
9. The Concept of a Person
10.Truths of Reason
References
Index

Citation preview

Knowledge and Justification

iCnowledge and Justification John L. Pollock Princeton University Press PRINCETON, NEW JERSEY

COPYRIGliT

© 1974 BY PRINCETON UNIVERSlfY PRESS Published by Princeton University Press Princeton and London

A II Rights Reserl'ed Library of Congress Cataloging in Publication Data will be found on the last printed page of this book Printed in the UNITED STATES OF AMERICA by Princeton University Press, Princeton, New Jersey

To HERBERT FEIGL who introduced me to philosophy

Preface

ONE of the most firmly entrenched beliefs of contemporary philosophy is that the only way to analyze a concept is to state its truth conditions. In epistemology this has led to the search for reductive analyses. These are analyses that state the truth conditions of concepts in terms of the grounds we employ in ascribing the concepts to things. Thus we are led to phenomenalism and behaviorism, and their analogues in other areas of knowledge. But these attempts at reductive analysis have invariably failed, leaving epistemology shipwrecked on the shoals of a barren theory of conceptual analysis. The purpose of this book is to defend an alternative theory of conceptual analysis according to which concepts can be analyzed in terms of their justification conditions rather than their truth conditions. The first two chapters provide a theoretical justification for this alternative scheme of analysis, but the strongest possible argument in its favor must consist of actually carrying out the proposed analyses. Accordingly, the remaining chapters attempt to provide the analyses for a number of epistemologically problematic concepts, and in so doing solve a number of traditional epistemological problems. Only the reader can judge how successful this attempt has been. The epistemological theory I am concerned to defend has historical antecedents, although they are surprisingly meager. The one area in which it has been pursued vigorously is the philosophy of mind. In that area it is represented by the "criteriological theory" apparently stemming from Wittgenstein and further developed by Malcolm, Strawson, Shoemaker, and others. My position is also suggested by some of Carnap's remarks on the logical concept of probability, although the connection is rather

Preface tenuous. The only philosopher who has seriously defended this sort of view across a wide spectrum of philosophical problems is Roderick Chisholm, and I am probably closer to him than to any other philosopher. A terminological matter should be raised here. Being an unrepentant Platonist, I make free use of the terms "concept" and "proposition". I also tend to use the term "statement" interchangeably with "proposition". Some philosophers may be put off by this Platonistic terminology. However, in the first nine chapters of the book, that is all it is—terminology. A philosopher who prefers words and sentences to concepts and propositions can translate my proclamations into his terminology without loss. It is only in the final chapter that the Platonistic terminology gains substance. In that chapter I argue explicitly for a Platonistic view of concepts and propositions and argue that, contrary to popular opinion, conceptual analysis is not about language. The contemporary view that all we are doing in philosophy is analyzing language is a myth. Many parts of this book have grown out of journal articles. The entire book can be regarded as growing out of my article "Criteria and Our Knowledge of the Material World". Chapter 1 is a minor revision of "What Is an Epistemological Problem?". Chapter 2 contains material from "Perceptual Knowledge" and greatly revised material from "The Structure of Epistemic Justification". Chapters 3 and 4 are taken, with minor revision, from "Perceptual Knowledge". Some of the material in chapter 8 comes from "The Logic of Projectibility". Part of chapter 10 grows out of "Mathematical Proof". I am indebted to The Research Foundation of the State of New York for three summer fellowships which helped considerably in the early stages of writing this book, and to The American Council of Learned Societies for a fellowship for the spring and summer of 1972 which allowed me to bring the book to completion. I am also indebted more than I can say to my colleagues, friends, and students for helpful comments and criticism along the way. I am particularly indebted to Rolf Eberle, Keith Lehrer, and John Turk Saunders. JOHN L. POLLOCK

Rochester, New York November 1972

Contents

Preface

vii

One What Is an Epistemological Problem? 1. The Fundamental Problem of Epistemology 2. Reductive Analyses 3. A Theory of Meaning 3.1 The Analysis of Concepts 3.2 The Analysis of Statements 4. TheDemiseofTraditionalEpistemology

3 6 11 12 17 20

Two The Structure of Epistemic Justification 1. The Pyramidal Theory of Knowledge 2. Epistemologically Basic Propositions 2.1 The Existence of Epistemologically Basic Propositions 2.2 The Logical Status of Epistemologically Basic Propositions 2.3 Conclusions 3. Reasons 3.1 SomeElementaryObservations 3.2 LogicalandContingentReasons 3.3 ConclusiveandNonconclusiveLogicalReasons 3.4 Logically Good Reasons and Prima Facie Reasons 3.5 Logical Reasons Again 3.6 Conclusions

23 25 25 29 32 33 33 34 36 39 46 48

Three Theories of Perceptual Knowledge 1. 2. 3. 4. 5.

Introduction NaiveRealism Phenomenalism Scientific Realism Descriptivism

50 51 53 54 56

Coatents 6. 7. 8. 9. 10.

Direct Realism The Principle of Implicit Reasons Reconsideration of Direct Realism Reconsideration of Naive Realism Conclusions

58 60 64

66 70

Four Incorrigibility 1. 2. 3. 4.

Preliminaries A Demonstration of Incorrigibility The Electroencephalograph Argument Conclusions

71 75 77 78

Five Perceptual Attributes 1. Introduction 2. "There is something red before me" 2.1 A Perceptual Criterion 2.2 Other Reasons 2.3 Defeaters 3. "χ has a red surface facing me" 3.1 A Perceptual Criterion 3.2 OtherReasons 3.3 Defeaters 3.4 Objective Appearance 4. Nonspatial Perceptual Attributes 5. Spatial Attributes 5.1 Length 5.2 Volume and Area 5.3 Shape 5.4 Location 5.5 Phenomenal Attributes 6. Perception 7. Perceptual Attributes 8. The Concept of a Physical Thing

80 81

81

82 84 85 85 86 86 95 97 98 98

108 111

113 116 119 128 129

Six The Reidentification of Physical Things 1. Introduction 2. Spatio-Temporal Continuity 3. Reidentification 3.1 ContinualObservation 3.2 Discrete Observation 3.3 ASingleCriterion 3.4 Defeaters 4. Sortals and Composition 4.1 Fission and Fusion 4.2 Composition

134 140 142 143 149 153 155 157 157 158

4.2.1 Composition and identity 4.2.2 The analysis of composition 4.2.3 Examples of composition 4.3 Reidentification 4.4 Emergence and Inheritance 4.5 Conclusions

Seven Memory and Historical Knowledge

Contents 158 161 164 166 171 174

1. Introduction 2. Theories of Historical Knowledge 2.1 Naive Realism 2.2 Reductionism 2.3 Scientific Realism 2.4 Descriptivism 3. Two Kinds of Memory 3.1 Propositional Memory 3.2 Personal Memory 3.3 Reconsideration of Propositional Memory 4. An Analysis of Propositional Memory

175 177 177 179 179 180 181 182 183 188 196

1. The Problem of Induction 2. The Confirmation Relation 3. Classical Attempts to State the Principles of Induction 3.1 The Nicod Principle 3.2 Explanatory Induction 4. Enumerative Induction 5. A Theory of Projectibility 5.1 Nonprojectible Conditionals 5.2 The Definition of Projectibility 5.3 The Logic of Projectibility 5.3.1 Closure conditions for projectibility 5.3.2 Some conjectures on projectibility 5.3.3 Indirect confirmation 5.4 Basic Projectibility 6. Projectibility and Induction 7. Theories and the Hypothetico-Deductive Method

204 205

Eight Induction

Nine The Concept of a Person

1. Theories Regarding Our Knowledge of Other Minds 1.1 The Argument from Analogy 1.2 Scientific Realism 1.3 Behaviorism 1.4 The Criteriological Theory

208 208 213 218 226 226 229 231 231 234 237 239 241 241 249 250 253 254 255 xi

Contents 2. Our Knowledge of the P-States of Others 2.1 The Equivalence of the Argument from Analogy and the Criteriological Theory 2.2 Persons and Bodies 2.3 The Problem of Copersonality 2.4 How P-Concepts Are Possible 3. The Analysis of P-Concepts 3.1 Projectibility 3.2 Phenomenological States 3.2.1 Identifying phenomenological states with one another 3.2.2 Reidentifyingphenomenologicalstates 3.3 The Projectibility of Phenomenological Sameness 4. Personal Identity 4.1 First-Person Reidentification 4.2 Third-Person Reidentification 4.3 Exchange of Bodies 4.4 Borderline Cases of Personal Identity 5. The Concept of a Person

264 264 265 267 272 274 274 276 276 281 282 283 283 288 292 293 296

Ten TruthsofReason 1. Introduction 2. Logical Skepticism 3. Two Theories of A Priori Knowledge 3.1 LogicalIntuitionism 3.2 Reductionism 3.3 Conventionalism 4. Logical Intuitionism 4.1 Platonism 4.2 Logical Intuitions 4.3 The Phenomenology of Our Logical Intuitions 5. A Priori Truth and Implication 5.1 Implication 5.1.1 Immediateimplication 5.1.2 Implication 5.2 LogicalPossibility 5.3 APrioriTruth 5.4 Some Formal Results 6. Logical Necessity and Entailment 7. Conclusions

300 302 304 304 305 308 317 317 318 321 327 328 328 329 331 332 333 336 339

References

341

Index

347

Knowledge and Justification

Chapter One

What Is an Epistemological Problem? 1. The Fundamental Problem of Epistemology LET US begin by looking at an example of a classical epistemological problem: There is a book sitting on my desk in front of me. But, now, suppose I ask myself how I know that there is, or more generally, how I know that there is anything there at all (regardless of whether it is a book). A sensible answer to this question would be, "Because I see it." We know that there are material objects around us because we see them, feel them, hear them, etc. And the statement that we see something, or feel it, or hear it, logically entails that it is there to be seen, or felt, or heard. But now, I say that I see something (a book) there before me on my desk, but how do I know that I do? Mightn't I be hallucinating, or seeing an after-image, or witnessing some sort of cleverly constructed optical illusion? My experience might be exactly the same as when I really am seeing a book, and yet there might not be any book there, because I am hallucinating. Generalizing this, how do we know we ever perceive the things we think we do? Mightn't we always be hallucinating? As it has sometimes been denied that it is even meaningful to suppose we might always be hallucinating, let us make this possibility more concrete. Suppose that a group of psychologists, biophysicists, and neurologists have constructed an adequate explanation of the neurophysiology of perception, and to test their explanation they take a subject from birth and wire him into a computer which directly stimulates his brain in such a way as to give a coherent, but completely false, sequence of sensations. In the subject's own mind he would seem to live out a completely normal life, growing up, making friends, going to school, getting a job,

What Is an Epistemological Problem?

marrying and raising a family, etc. And yet all those years he was really sealed into an experimental apparatus in which he was fed intravenously and never had any contact with the outside world. It is true that in the present state of neurophysiology this could not be done, but it is certainly a meaningful hypothesis and a logical possibility. Now, how do I know that I am not in the position of the subject of the above experiment? Perhaps a group of scientists have me hooked into such a computer, and all of the experiences that I think I have had since birth are really figments of the computer. How can I possibly know that this is not the case? It seems that any reason I can have for thinking that I am not hooked into such a computer must be either a reason for thinking that such a hypothesis is logically impossible, or else an empirical reason, arrived at in­ ductively, for thinking that it is false as a matter of fact. But it is hard to see how this skeptical hypothesis can be logically im­ possible (it seems to make perfectly good sense—we know what it would be like for someone to be wired into a computer), and it seems that in order to have inductive evidence for an empirical reason I would already have to be able to rely upon some of my perceptions—which I cannot do without simply begging the question against the skeptical hypothesis. How then can I know that the skeptical hypothesis is false? Faced with this sort of argument, one might be tempted to conclude that the skeptic is right—we really don't know the things we think we know. But such a conclusion flics in the face of com­ mon sense. There are many things that I know: I know that there is a desk before me, that I am holding a pen and writing on a piece of paper. I know that the walls of my study are lined with books, and that it is raining outside. All of these things I am certain about. It would be ridiculous to conclude that it is in principle impossible for me to know them. That an argument " P u . . . , P n ) therefore Q " is valid does not establish that its conclusion is true. It merely establishes that if the premises are true then the conclusion is true. Or better, it establishes that at least one of the propositions Pu . . . , Pn, Q is false. The argument does not by itself determine which is false. A skeptical argument proceeds from prima facie reasonable prem­ ises to the conclusion that we do not know things that we are quite certain we do know. But all that such an argument establishes

1. The Fundamental Problem of Epistemology is that either one of the premises P1, . . . , Pn is false or that proposition Q (that we do have the sort of knowledge we think we do) is false. In deciding which of these propositions to reject, all we can do is seize upon the one we are least certain about. But we will never be as certain about the premises of a skeptical argument as we are that we do have knowledge. So it will always be more reasonable to reject one of the premises than to accept the skeptical conclusion. A skeptical argument can only be construed as a reductio ad absurdum of its premises. There must be something wrong with any skeptical argument. Presented with such an argument, what we must decide is which premise to reject. Skeptical arguments generate epistemological problems. Apparently reasonable assumptions lead to the conclusion that knowledge of a certain sort (e.g., knowledge of the physical world, or knowledge of other minds) is impossible. Faced with such an argument, our task is to explain how knowledge is possible. The problem is not to show that knowledge is possible; that much we can take for granted. What we must do is find the hole in the skeptical argument that makes it possible for us to have the knowledge we do. The problems of epistemology are problems of how we can possibly know certain kinds of things that we claim to know or customarily think we know. In general, given a statement P, we can ask, "How do you know that PT' This is the general form of an epistemological problem. The question "How do you know that P?" is a challenge—a demand for justification. The task of the epistemologist is to explain how it is possible for us to know that P, i.e., to explain what justifies us in believing the things we do.1 Returning for a moment to the skeptical argument with which this chapter began, we can be confident that it proceeds from a false premise. Its conclusion, which is that knowledge of the physical world is impossible, is certainly mistaken. The task of the epistemologist is to find the false premise. This is not a task that can be undertaken at this point. Preliminary groundwork is 1 To forestall misunderstanding, let me hasten to say that the task I am undertaking is not that of investigating knowledge per se but merely that of investigating epistemic justification. Gettier [1963] has made it painfully obvious that knowledge requires more than justified true belief, but in this book I will not try to say what more is required. The task I have set myself is not to determine when we have knowledge but is the more funda­ mental one of determining when our beliefs are justified.

What Is an Epistemological Problem?

necessary. We will return to this skeptical argument in Chapter Two, at which time it will be possible to pinpoint the error.

2. ReductiveAnalyses Now let us turn to a second skeptical argument, around which the contents of this book will be organized. The development of this argument is rather involved, occupying this entire section. At the end of the section, the argument will be summarized. Our knowledge can be separated into areas according to subject matter. These areas will include knowledge of the physical world, knowledge of the past, knowledge of contingent general truths, knowledge of other minds, a priori knowledge, and possibly knowledge of moral truths. The significance of these areas is that each has associated with it a characteristic source of knowledge. For example, the source of our knowledge of the physical world is perception. This is not to say that the only way to know that a physical object has a certain property is by perceiving the object. There are other ways, e.g., we may remember that it does, or we may be told that it does. But these other ways are all parasitic on perception. If we could not acquire knowledge of the physical world through perception, we could not acquire it in any of these other ways either. Analogously, the source of our knowledge of the past is memory; the source of our knowledge of contingent general truths is induction; the source of our knowledge of other minds is the behavior of other bodies. It is not clear just what are sources of knowledge either of a priori truths or of moral truths. Given a statement P, let us call the conditions under which one would be justified in believing-that-P the justification conditions of the statment P. We can distinguish between two problems. The first is to state the justification conditions for the propositions in each of our areas of knowledge, and the second is to prove that those are the justification conditions. These two problems are not unrelated, but they are distinct problems. The second has generally interested epistemologists more than the first. Epistemologists have usually been content to give only a very rough description of the justification conditions of statements, and then have gone ahead to try to prove that those are the justification conditions. For example, the classical Problem of Induction is one of justifying induction as a way of learning the truth of universal

2. Reductive Analyses generalizations. Although they've never been very clear about just what those grounds are, few philosophers doubt that we do base knowledge claims on inductive grounds. But what is demanded is a proof that we are justified in doing so. And similarly, the Problem of Perception is the problem of explaining how we can justify basing knowledge claims about physical objects on sense perception. It is not doubted that we do in fact base them somehow on sense perception, but what is wanted is a proof that we are justified in doing so. The fundamental problem of epistemology is to explain what it is that justifies us in making the kinds of knowledge claims that we do customarily make. This problem has traditionally been construed as requiring a justification for our basing knowledge claims on the grounds on which we do in fact base them (a proof that what we suppose to be the justification conditions really are the justification conditions). In other words, it has been identified with the second of the above two problems. On the face of it, there seems to be a very good reason why we should, in principle, be able to give a proof of the sort desired. If we cannot establish any connection between one state of affairs and another, then we cannot be justified in making claims about the one state of affairs on the basis of the other. Thus if we cannot justify our customary grounds for knowledge claims, then we cannot take them as justifying our claims to knowledge. But if we cannot take them as justifying those claims, then they do not justify those claims, and so they are not really good grounds at all. Therefore, unless we can, in principle, give a proof of the sort desired, we are led to skepticism. How might we set about justifying our basing knowledge claims on some particular source (such as perception)? It seems that there can be only two ways in which this might be done. We could either justify it inductively, showing that it does in. fact tend to lead to true knowledge claims, or else we could justify it logically, showing that there is some sort of logical connection between the source and the knowledge of which it is a source. But an inductive justification is impossible. We could only inductively justify a source of knowledge if we had independent access to both the source and the knowledge of which it is a source, and then could compare them and see that there is a correlation. But we do not have independent access to the knowledge that these sources are supposed

What Is an Epistemological Problem?

to provide. They constitute the sources of this knowledge. For example, we do not have access to the physical world except through perception, and so there is no way to compare the physical world with perception to see that perception is a reliable guide un­ less we beg the question and assume in the beginning that it is. It seems then that the only way to justify a source of knowledge is by establishing some sort of logical connection between the source and the knowledge it is supposed to give us. A logical connection must arise from the meanings of the concepts or statements in­ volved in the knowledge claims. And (and here is the step which I shall deny) it has traditionally been supposed that the only way to analyze the meaning of a statement or concept is to give its truth conditions—to say what conditions must be satisfied in order for the statement to be true or for the concept to be correctly ascribable to an object.2 Furthermore, not just any statement of truth conditions will suffice. Starting from the truth conditions of a statement, we could never establish a logical connection between that statement and the source of knowledge which is supposed to yield the statement unless those truth conditions were stated in terms of the same concepts as are used in describing the source. Thus, for example, we could never establish a logical connection between perception and statements about the material world unless we could state the truth conditions of the latter in terms of the concepts used in describing perception. An analysis of the truth conditions of a statement in terms of the concepts used in describing the source of our knowledge of that kind of statement is what philosophers have called a reductive analysis. Since Descartes, epistemologists have been concerned almost exclusively with giving reductive analyses of statements, and we now see why. Given assumptions that philosophers have traditionally accepted, it follows that the only way to prove that the purported justification conditions of a statement really are the justification conditions is by giving a reductive analysis of the statement in which the truth conditions are stated in terms of the same concepts as the justification conditions. Philosophers have commonly supposed that they know more or less how we are justified in making the kinds of knowledge claims 2We have inherited this assumption most recently from logical atomism and logical positivism, but it is not new. It is really the same sort of thing that Locke, Berkeley, and Hume, and' for that matter Socrates, were doing.

2. Reductive Analyses we do. Although they have not generally been able to state the justification conditions precisely, they felt that they could at least pick out the general sources of our knowledge in different areas. For example, our knowledge of the physical world comes from sense perception. And our knowledge of other minds comes from people's behavior. And our knowledge of right and wrong (if we can properly speak of "knowledge" here) comes from nonmoral states of affairs in the world. Thus if we are to justify these sources of knowledge, it seems we must seek reductive analyses of these statements in terms of these sources. Working within this tradi­ tional framework, phenomenalism becomes the only possible theory of our knowledge of the material world, behaviorism be­ comes the only possible theory of our knowledge of other minds,3 and some form of naturalistic ethics becomes the only possible theory of our knowledge of moral truth. It seems that in order to justify sources of knowledge we are driven inexorably to reductive analyses. This appears to be the only way to derive the justification conditions of statements from the meanings of those statements. And it seems that if the justifica­ tion conditions are not derivable from the meanings of the state­ ments—if there really is no logical connection between them— then they cannot be the justification conditions, because if we cannot justify our sources of knowledge, then they do not justify our claims to knowledge and so are not really sources of knowledge at all. We seem forced to conclude that we must have either reductive analyses or skepticism. They are the only two pos­ sibilities. Let us ask then whether reductive analyses are always possible. Unfortunately, the answer seems to be "No". For example, con­ sider our knowledge of the physical world. A reductive analysis there would take the form of phenomenalism. I think the strongest argument against phenomenalism is that from "perceptual rela­ tivity". According to phenomenalism a statement about physical objects is to be analyzed as a (perhaps infinite) conjunction of s This is oversimplifying a bit in the case of behaviorism, because the possibility of an inductive justification of our source of knowledge has been maintained by the supporters of the argument from analogy. But if we can agree with those who maintain that the argument from analogy will not work, and thus that an inductive justification is impossible, then it seems to follow that behaviorism is the only possible theory.

What Is an Epistemological Problem?

statements about experience. Then the physical object statement must entail each of those statements about experience. But a physical object statement cannot entail any statement about experience: given any physical object statement and any statement about experience, we can imagine circumstances in which the physical object statement is true, and yet due to malfunction of certain organs, or hallucination, the statement about experience is false. For example, we can recall the example of the man who is wired into a computer that directly stimulates his brain. This is done in such a way that his experience is entirely independent of what is actually going on around him, so that his experience is always delusive. Given any proposed entailment between a material object statement and a statement about his experience, we can always program the computer in such a way as to make the latter statement false even when the former is true. Consequently there is no entailment. Thus it seems that the search for reductive analyses that will solve the problem of our knowledge of the physical world is stymied. Nor is this the only place in which the attempt to find reductive analyses seems to have failed. Consider our knowledge of other minds. Here the reductive analysts are the behaviorists who try to reduce statements about persons to statements about material objects. But behaviorism does not seem to work,' and so reductive analyses do not offer a solution to the problem. Consider the philosophy of logic. Quine and his followers have argued, in effect, that no reductive analyses can be given of such logical concepts as analyticity, implication, and logical necessity, and on this basis they have actually been led to embrace a kind of logical skepticism wherein they deny the very existence of these concepts. We have a corresponding problem in ethics. Reductive analyses in ethics take the form of naturalistic ethical theories in which the meaning of an ethical statement is analyzed in terms of statements about the physical world. But the apparent existence of the Naturalistic Fallacy vitiates such analyses. In all of these areas the search for reductive analyses has been fruitless. Thus the traditional epistemologist finds himself forced to the conclusion that either reductive analyses can be given in the various areas of knowledge or else skepticism is true. And he cannot find any reductive analyses. 4

See Chapter Nine.

3. A Theory of Meaning The above skeptical argument has been rather long and involved, so let me lay it out explicitly step by step: 1. If we cannot establish a connection between one state of affairs and another, then we cannot be justified in making claims about the one state of affairs on the basis of the other. 2. Thus if we cannot establish a connection between our customary sources of knowledge and the states of affairs of which they are supposed to yield knowledge, then they are not sources of knowledge after all. 3. It is impossible to establish such a connection inductively without begging the question. 4. Thus the only way to justify a source of knowledge is to establish a logical connection between it and the states of affairs of which it is supposed to yield knowledge. 5. A logical connection must arise out of the meanings of the knowledge claims. 6. The only way to analyze the meaning of a statement is to give its truth conditions. 7. Thus the only way to justify a source of knowledge is to analyze the truth conditions of the knowledge claim in terms of the concepts used in describing the source of knowledge, i.e., to give a reductive analysis. 8. It is impossible to give reductive analyses. 9. Therefore our sources of knowledge cannot be justified and hence are not sources of knowledge at all. The conclusion of this argument is certainly false, so one of the steps of the argument must fail. The philosopher's problem is to discover which. 3. ATheoryofMeaning The solution to our skeptical dilemma is to deny that the meaning of a statement must always be given by stating its truth conditions, i.e., to deny step 6 of the above argument. It will be argued that often the meaning of a statement can be given in another way—by stating its justification conditions. Or more precisely, often the meaning of a statement is uniquely determined by the justification conditions of it and its denial. The more I think about this, the more obvious it seems to me and the less in need of argument.

What Is an Epistemological Problem? Despite the exhaustive search for truth condition analyses which has occupied philosophy for many years, such analyses are distin guished by their rarity. It is the truly exceptional statement or concept for which truth condition analyses have actually been found. Stop and think how many even simple concepts can literally be "defined". To learn the meaning of a concept is certainly not to learn its "definition". It is to learn how to use it, which is to learn how to make justifiable assertions involving it. Thus it seems to me inescapable that the meaning of a concept is determined by its justification conditions. This seems to me obvious, but I realize it does not seem obvious to those steeped in the tradition of analytic philosophy, so I will now endeavor to give a detailed argument to establish what I think is really obvious. My initial thesis is that often, as a matter of logic, the meaning of a statement is uniquely determined by the justification conditions of it and its denial. This thesis will be defended in two steps. First it will be shown that it holds for statements of the simple form "x is a φ", and then it will be concluded from this that the thesis holds in general. 3.1 The Analysis of Concepts Rather than talk directly about the statement "Λ: is a 9", it proves convenient to talk about "the concept of a p". By "the justification conditions of the concept of a 95" will be meant the justification conditions of the statement "x is a Q ) . T h e c o n j u n c t i v e b e l i e f - t h a t - ( P a n d ( P Z ) Q ) ) i s a logical reason for S to believe that Q. Consequently, although there are these two distinct classes of good reasons—logical and contingent—contingent reasons can always be reduced to logical reasons. Therefore, justification can be thought of as proceeding exclusively in terms of logical reasons. 3.3 Conclusive and Nonconclusive Logical Reasons Simple entailments are logical reasons. It is tempting to sup­ pose that all entailments are logical reasons. But this will not do. If we consider any fairly complex entailment, one might be justi­ fied in believing the antecedent but, not knowing that the anteced­ ent does entail the consequent, not be justified in believing the consequent. For example, the Axiom of Choice entails Zorn's Lemma, but this is not an entailment that one could be expccted to see without proof. Thus one might justifiably believe that the Axiom of Choice is true but not see that it entails Zorn's Lemma, and so not be justified in believing that Zorn's Lemma is true. In order for an entailment to provide a good reason, one must be justified in believing that there is such an entailment. In the case of simple entailments, we can see just by considering the meanings of the statements that the one entails the other, and so such en­ tailments are always good reasons for anyone who understands the statements involved. But there are more complex entailments that can only be known as the result of giving a demonstration, and these entailments do not provide us with good reasons until we have given the requisite demonstrations. To give such a dem-

3. Reasons onstration is to acquire an independent reason for believing that if the antecedent is true then the consequent is true. Hence, although after the demonstrations have been given the entailments provide good reasons, they are not logical reasons. They are only contingent reasons. We have isolated one class of logical reasons, which we can call "conclusive" reasons. Let us say that: (3.2) The statement-that-P is a conclusive reason for S to believe that Q, iff the statement-that-P entails and is a logical reason for S to believe that Q. Conclusive reasons are logical reasons. Do conclusive reasons exhaust the class of logical reasons? It is astonishing how often philosophers have thought that they do. For example, Judith Jarvis Thompson [1965], p. 292, writes, "One thing that seems plain is that if 'This is S' does not imply 'This is P' it will always be at best a matter of fact, to be established by investigation, that the first is a reason for the second."12 But there seem to be clear counterexamples to the supposition that all logical reasons are conclusive. Consider, for example, induction. A class of singular statements P1, . . . , Pn can provide inductive grounds for an unrestricted generalization Q, and thus the conjunction (P1 & . . . & Pn) of those singular statements is a good reason for the general statement Q. The conjunction of the singular statements clearly does not entail the generalization, and yet it constitutes a logical reason for believing it. This can be shown as follows. If (P 1 & . . . & P n ) constituted merely a contingent reason for believing that Q, then a person would have to have an independent reason for believing that [(Pi & . . . & Pn) Z> Q] before the conjunction could be a reason for believing that Q. There are only two plausible alternatives for what such an independent reason might be. First, it might be an inductive reason. But then we could ask the same question with regard to that inductive reason. This appeal to further inductive reasons can only go on for finitely many steps (recall that an infinite regress of justification is impossible); eventually there has to be a stopping point which must be justified on some other basis in order to get the whole sequence of inductive justifications started in the first place. This stopping point will be a conditional of this same form which is justified in some other way. This brings us to 12

See also Russell [1948].

The Structure of Epistemic Justification

the second alternative: this conditional might be entailed by some general principle which we might call a principle of the Uniformity of Nature. Philosophers have often looked for some such principle to justify induction. But even if such a principle could be found, which seems exceedingly unlikely, we could still ask what justifies us in believing it. Again, there are two possibilities. First, the principle might be a truth of logic. But in that case the conditional [(P1 & . . . & Pn) Z> Q], being entailed by the general principle, would also be a truth of logic. And this would require that the conjunction of singular statements (P1 & . . . & Pn) entail the gen­ eral statement Q, which is impossible. Suppose instead that the general principle is not a truth of logic. Then we can ask what logical reasons there are for believing it. And on the supposition that all logical reasons are conclusive, this amounts to asking what justified beliefs entail it. These beliefs constitute a set Γ. The beliefs in Γ must be justified without appeal to the general principle, so they cannot include any beliefs justified inductively. Consequently, any general beliefs in Γ must be truths of logic. But if a set of statements some of which are truths of logic entails another state­ ment, then the set of statements that results from deleting the truths of logic also entails the other statement. Consequently, any general beliefs in Γ can be omitted with the result that the remain­ ing beliefs still entail the principle of the Uniformity of Nature. The remaining beliefs in Γ must constitute a class of singular statements R u . . . , R m . But then if the inductive grounds P u . . . , Pn together with the principle of the Uniformity of Nature entailed Q, the inductive grounds Pi, . . . , Pn together with the additional singular statements R u . . . , R m would also have to entail Q (because the latter entail the principle of the Uniformity of Nature). And then again we would have a general statement being entailed by a finite conjunction of singular statements, which is impossible. Thus induction cannot be accounted for in terms of conclusive reasons, and so cannot itself be a contingent reason. But it cannot be denied that induction provides us with good rea­ sons. It must be concluded that induction constitutes one example of a nonconclusive logical reason. And it will be argued throughout this book that there are many other important examples of such reasons.18 13This of course should not be taken as implying that induction and other nonconclusive logical reasons cannot be "reduced" to conclusive rea-

3. Reasons

The supposition that all logical reasons are conclusive is intimately connected with the traditional hope that all epistemological problems can be solved by giving reductive analyses. Beginning with the pyramidal theory of knowledge, philosophers have been enamored of the idea that each statement in the pyramid is either epistemologically basic or else can be expressed as a logical construction of statements falling in lower tiers of the pyramid. This would provide us with reductive analyses of all statements in terms ultimately of epistemologically basic statements. On the supposition that such reductive analyses were always possible, it became plausible to suppose that all logical reasons were conclusive. One could think of the process of justifying a belief as proceeding upward through successively higher tiers of the pyramid in such a way that each statement was a logical construction of statements below it and thus entailed by them. Then ultimately a justified belief would be entailed by the epistemologically basic statements at the bottom of the pyramid. But in recent years it has come to seem quite unlikely that this enterprise of seeking reductive analyses can be successful. It will be established in subsequent chapters of this book that reductive analyses cannot generally be given. It becomes accordingly less likely that justification can always be viewed in terms of entailment. This makes it plausible to suppose that there are many logical reasons that are not conclusive. One of the principal objectives of this book will be to establish that this is indeed the case in most of the philosophically problematic areas of knowledge. It is precisely the unacknowledged presence of nonconclusive logical reasons in those areas that has made the areas philosophically problematic. 3.4 Logically Good Reasons and Prima Facie Reasons Logical reasons that are not conclusive are particularly interesting because they have been largely overlooked by philosophers bent upon finding reductive analyses. In them lies the key to a number of stubborn epistemological problems. Let us call such reasons logically good reasons. Whereas conclusive reasons guarantee truth, logically good reasons only guarantee justification. sons, utilizing the techniques of section 3.2, by conjoining them with a condi­ tional stating that if the reason is true then the conclusion is true. But the point i9 that these reasons are logical reasons all by themselves without con­ joining them with such a conditional.

The Structure of Epistemic Justification

Induction provides us with one example of a logically good reason. The inductive grounds for a conclusion do not constitute a con­ clusive reason for believing the conclusion, but, as was argued above, they do constitute a logical reason. Many logically good reasons have a certain kind of structure which makes it reasonable to call them prima facie reasons. A prima facie reason is a logical reason that is defeasible. In other words, a prima facie reason is a reason that by itself would be a good reason for believing something, and would ensure justifica­ tion, but may cease to be a good reason when taken together with some additional beliefs. Induction provides us with one example of a prima facie reason. An inductive reason is a logical reason, and it is clearly defeasible. An inductive reason for accepting a generalization can be defeated on at least two grounds. First, no matter how strong the initial inductive evidence for the generaliza­ tion, if further investigation reveals a counterexample then the original reason ceases to be a good reason. Second, if it is dis­ covered that the sample on which the original generalization was based was not a fair sample, this will make the initial reason no longer a good reason even though it was a good reason until this was discovered. Another, perhaps more interesting, example concerns perceptual judgments. For now, I will consider only one instance of a per­ ceptual judgment—a person's judgment that something is red on the basis of its looking red to him—but it seems clear that the conclusions drawn in connection with this one example have rather broad application to perceptual judgments in general. In the next chapter it will be demonstrated that "x looks red to S" is a logical (in fact, a prima facie) reason for S to believe that χ is red, so let us assume for now that this is true. This should seem plausible without much argument. It seems indisputable that there must be some sort of logical connection between "x looks red to 5" and "χ is red". It is not just an accident that red things tend to look red to people. This vague intuition is fortified by the ob­ servation that to suppose otherwise would make it impossible for us to ever know that anything is red. If we were to suppose that the connection between something's looking red to us and its actually being red is only a contingent connection, then the only way we could ever establish the connection is inductively. But we could never establish inductively that things that look red to us

3. Reasoris

tend to be red, because in order to do that we would have to be able to tell independently what things are red, and the only way we have of doing that is in terms of what things look red to us, which would beg the question. Therefore, if the connection were merely contingent, then knowledge of red objects would be impossible. But knowledge of red objects is possible, so the connection cannot be contingent; it must be a logical connection of some sort.14 It was the hope of phenomenalism that this logical connection could be explained entirely in terms of conclusive reasons. It was hoped that the meaning of "x is red" could be analyzed in such a way that x's looking red under certain specifiable phenomenological conditions would logically entail that χ is red. We saw in Chapter One that no such analysis can be given. Consequently this logical connection cannot be explained entirely in terms of conclusive reasons. But given that it is a logical connection, it must be explained in terms of some sort of logical reasons, and hence these reasons must be logically good reasons. Can we describe the logical connection between "χ looks red to S" and "x is red" in such a way as to elicit the structure of the logically good reasons involved? Ordinarily, when I can see an object clearly, and have no reason for supposing that there is something wrong with my eyes, or that there are strange lights playing on the object, or anything of that sort, I unhesitatingly judge that the object is red if it looks red to me. If I have no beliefs about χ other than that it looks red to me, then I am justified in thinking that it is red, and this is so simply by virtue of the con cepts "red" and "looks red". But if I do have certain other beliefs, my belief that χ looks red to me may not justify me in believing that χ is red. For example, I may believe that there are red lights shining on χ and that in the daylight it looks white. If I had those beliefs, the simple fact that χ looked red to me would not justify me in believing that χ was red. Thus the belief that χ looks red to me is a defeasible logically good reason for me to think that χ is red, i.e., it is a prima facie reason. A proposition which, when added to a prima facie reason, defeats the justification, is called a defeater. We can define this precisely as follows: 14 This

argument will be strengthened considerably in the next chapter.

The Structure of Epistemic Justification (3.3) If P is a logical reason for S to believe that Q , then R is a defeater for this reason iff the conjunction ( P & R ) is not a logical reason for S to believe that Q . For example, " x is illuminated by red lights, and red lights often make an object look red when it is not" is a defeater for "JC looks red to S" as a logical reason for S to believe that JC is red. A prima facie reason is simply a logical reason for which there exist def eaters: (3.4) P is a prima facie reason for S to believe-that-Q iff (1) P is a logical reason for 5 to believe that Q, (2) there is a proposition R which is a defeater for P as a reason for S to believe that Q, and (3) R is consistent with P. 15 Now let us examine more closely the nature of dcfcaters. .To say that P is a prima facie reason for S to believe-that-0 is to say that in the absence of any other information S is justified in believing that it would not be true that P unless it were true that Q. Let us symbolize this subjunctive conditional as "P => Q'\ Then P is a prima facie reason for S to believe-that-β iff S is prima facie justified (in the sense of section 2.2) in believing-that-(P => Q). A defeater must be a reason for thinking that this conditional is f a l s e , i . e . , a r e a s o n f o r S t o b e l i e v e t h a t —- ( P — > Q ) . There are two ways in which a proposition can be a reason for denying a conditional, and accordingly there are two kinds of defeaters. First3 if P is a prima facie reason for S to believe lhat Q , then any reason for S to believe that Q is false (even though P is true) is a defeater. Such a defeater defeats the conditional truthfunctionally. Let us call these type I defeaters. For example, '"Jones told me that χ is not red, and Jones is generally reliable" would be a type I defeater for "x looks red to me" as a prima facie reason for me to believe that χ is red. Analogously, "That crow is not black" would be a type I defeater for an inductive reason for thinking that all crows are black. The second kind of defeater attacks the connection between P and Q rather than attacking Q directly. For example, although the 15 Clause 3 is required because, even in the case when P entails Q . ( P & — P) is not a logical reason for believing Q, and hence — P is a defeater. This sort of defeater must be ruled out, or all logical reasons would be prima facie.

3. Reasons belief that there are red lights shining on χ is not a reason for thinking that χ is not red, it is nevertheless sufficient to prevent the prima facie reason that χ looks red to me from justifying the belief that χ is red, and hence it is a defeater. This second kind of defeater is, roughly speaking, a reason for thinking that, under these circum stances, knowing-that-P is not a good way to find out whether Q. For example, if there are red lights shining on x, then knowing that χ looks red to me is not a good way to find out whether χ is red, because the red lights can make a white object look red. Let us call these type II defeaters. A type II defeater is any reason for believing that are members of a time worm whose perceptual attributes undergo no discontinuous changes" is a prima facie reason for judging that X ==» Y. The defeaters for this prima facie reason are inductive. Reidentifying on the basis of this and other reasons yet to be elicited, we discover that under specifiable circumstances certain other attributes (such as the presence of personal documents in a secret compartment) are relatively stable, and hence the failure of these attributes to be preserved in a time worm becomes an inductive reason for denying substantiality. We have seen that continuity of appearance is a prima facie reason for reidentifying. Is it also a criterion? In other words, does the existence of a discontinuity of appearance constitute a prima facie reason for denying temporal identity? Somewhat surprisingly, it does not. For example, if the color of my typewriter suddenly changed to a fluorescent orange, but it remained otherwise the same, I would have no difficulty in reidentifying it. I would rcidentify it on the basis of its other perceptual attributes remaining the same or changing only continuously. It may even be possible to reindentify an object if all of its perceptual attributes change discontinuously at the same instant, provided the changes are not too great. For example, suppose we are viewing an oriental figurine on a table. There is a flash of light, and suddenly the figurine has moved several inches to the right, has shrunk slightly and become

3. Reidentification somewhat misshapen, its color has darkened a bit, etc. Even given all of these simultaneous discontinuous changes, there would be little tendency to deny that it is the same figurine. Clearly, the reidentification is made on the basis of the similarity of the appearance of the figurine before and after the flash. It might be supposed that when we reidentify in the face of discontinuous changes in appearance, we are doing so on the basis of an inductive reason rather than a logical reason. This would require us to discover inductively on the basis of the case of continuous change that similarity of appearance is indicative of temporal identity, and then apply this inductive conclusion, to the case of discontinuous change. But a simple argument suffices to show that this cannot be the case. It is at least logically possible that all change might be discontinuous. It is quite possible that we might discover that all change takes place in terms of quantum jumps. Or if it is objected that quantum jumps are so small that at the macroscopic level such changes in appearance should still be regarded as continuous, we might suppose instead that all change in appearance takes place by means of very rapid discontinuous overcorrections. High-speed photography might reveal to us that, for example, when the color of a cooling piece of iron appears to be varying slowly and continuously from red to gray over a period of three minutes, the color change actually occurs in large discontinuous steps. When the object appears to have darkened just slightly, what has actually occurred is that it has abruptly become much darker, held that state for 10-e seconds, and then discontinuously changed back to almost the original color and held that for IO-3 seconds; then the process repeats with the iron coming out a slightly darker color than before, and so on for approximately 180,000 times until finally the color is that of cold iron. We might discover that all changes in perceptual attributes proceed by this kind of discontinuous overchange and correction, so that there is no such thing as true continuous change of perceptual attributes. In this case it would not be possible to inductively conclude on the basis of the case of continuous change that certain things are indicative of temporal identity because there would be no continuous change. But, in point of fact, this would not hamper us in our ability to reidentify physical things. Consequently, the reason to which we

The Reidentification of Physical Things appeal in the case of discontinuous change must be a logical reason, What does that reason amount to? We reidentify across a discontinuous change in appearance on the basis of the appearance of the object before the discontinuous change being sufficiently like the appearance of the object after the change. There is a "critical degree of similarity" such that if the objects bear at least that degree of similarity we have a reason for reidentifying them, and if they are not that similar we do not have a reason. So we have: (3.2)

ii X

and Y are substantial time worms, Y beginning when X ends, and the final stages of X bear at least a critical similarity to the initial stages of Y" is a prima facie reason for thinking that the concatenation of X and Y is substantial, the strength of the reason being a function of the degree of similarity.

Notice that principle 3.1 is just a special case of principle 3.2. In fact, it is just the limiting case in which the degree of similarity is total throughout the time worm. Consequently, we do not need principle 3.1 as a special principle—it is a corollary of principle 3.2. What happens if more than one object contemporary with Y bears a critical similarity to Xl Clearly, we reidentify X with the object to which it is most similar. This comes immediately out of principles 3.2 and 1.5. According to 1.5 it is a necessary truth that

x^ y & y ^ z & C o n c r / z ' ) . D X ^ Z . Now suppose that Y and Z are two contemporary objects each temporally contiguous with and bearing a critical similarity to X, but suppose Y is more similar to X than Z is. By virtue of 3.2 we have a prima facie reason for thinking that X ^ Y, and we have a prima facie reason for thinking that X " for the counterfactual biconditional which is the conjunction of the two counterfactual conditionals, we have: (4.4) X is composed of Y iff (3/4) (3 B) ( 3 /) ( 3 y) : 3Γ = 7χΑιΧ & Y = ηχΒιΧ & φ ( Υ ) & [( 3 \x)BiX & ( 3 \x)AiX], To avoid any misunderstanding, let me emphasize that the biconditional in this analysis is simply a symbolization of the following English sentence: "If there were no unique A at location I, then

The Reidentification of Physical Things either there would be no unique B at location I or that B would not be ψ\ and if there were no unique B at location I or that B were not φ, there would be no unique A at location It is perhaps unfortunate that this analysis makes such thoroughgoing use of counterfactual conditionals. Those who abhor counterfactuals will abhor this analysis. But there is nothing to be done about this—the counterfactuals appear to be essential to the analysis. Although I see no way to prove this, the composition relation, as analyzed by 4.4, does not seem to be symmetric. For example, the argument that demonstrated the inadequacy of 4.2 fails for 4.4: If the A at I is composed of the B at I (under the attribute ψ), then letting φ* be "is composed of the unique B at we certainly h a v e ( 3 \x)Btx]. But the converse fails. Without the additional assumption that the B at I is φ, there being a unique B at I is no guarantee of the existence of a unique A at I which is composed of that B. As one would expect, the composition relation is transitive: (4.5) If the A at ? is composed of the B at I, and the B at I is composed of the C at I, then the A at ί is composed of t h e C a t I. For example, if the car is composed of a certain piece of plastic, and the piece of plastic is composed of a certain collection of molecules, then the car is composed of the collection of moiccules. That this is true in general can be seen by letting the attribute φ which connects A and C be "constitutes a B which constitutes an A".

4.2.3 Examples of composition. Given analysis 4.4, we can now verify that the examples proposed in section 4.2.1 are indeed examples of composition. It is immediate that this is the relation between a statue and the lump of clay from which it is molded. The second case is more problematic. It seems to me intuitively that the relation between a club and the set of its members is the same as the relation in the other examples of composition. But it is not clear how to apply 4.4 to this case, because it is not immediately clear what is to count as the location of a club and the set of its members. I propose that we take the location both of the club and of

4. Sortals and Composition the set of its members to be the total volume of space occupied by its members. Given this convention, we can apply 4.4. For the club to exist is for the set of its members to be non-empty, so letting ψ be "is non-empty", it follows from 4.4 that a club is composed of the set of its members. Turning to the next case, it has frequently been maintained that the relation between a body of water and the corresponding set of H 2O molecules is one of identity. There are several reasons why this cannot be. First, the criteria for reidentification for bodies of water and for sets of molecules are quite different. The identity of a set is determined by the identity of its members. Changing one molecule is sufficient to change the identity of the set of molecules, but it is certainly not sufficient to change the identity of the body of water. Second, it verges on a category mistake to say that a body of water is a set of anything. The body of water is a physical thing, but a set is an abstract entity. Connected with this is the fact that the body of water and the set of molecules have different attributes. For example, the set of molecules may have cardinality 10®, but the body of water has no cardinality. Thus the body of water cannot be identified with the set of molecules. Instead, the relation is one of composition. That the body of water is composed of the set of H2O molecules follows immediately from 4.4, in this case letting ψ be "has as members all of the molecules at location Γ'. β Of course, this does not preclude that the body of water and the set of H 2O molecules also stand in some stronger relation to one another, but I think it does remove the temptation to suppose they do. The one being composed of the other suffices to explain the intimate connection we feel between the two things. Perhaps what tempts us most to suppose that the body of water is to be identified with the set of H 2O molecules is that we ordinarily say "Water is H2O". But we also say "That piece of clay is a statue", where the relation is even more obviously not one of identity.7 It must be recognized that "is" can mean more than one thing. There is an "is" of composition just as there is an "is" of identity. β Perhaps we should add something about temperature and density here too. T It is noteworthy in this connection that although we say "The piece of clay is a statue", we do not say "The statue is a piece of clay". The rela­ tion expressed by "is" is not symmetric.

The Reidentification of Physical Things Turning to lightning, it is certainly composed of an electrical discharge, but this does not automatically preclude the lightning's being identical to the electrical discharge, 8 and most philosophers have supposed that this is an identity. But there are several reasons why it cannot be. First, the criteria of reidentification for lightning and electrical discharges differ. As a matter of fact, the electrical discharge generally begins (comes into existence) several minutes before the lightning and builds up slowly. Furthermore, a single discharge could give rise to two consecutive bolts of lightning if it had an interval of very low discharge in the middle. Second, the lightning has attributes not possessed by the discharge. The lightning is yellow and pointed, but an electrical discharge cannot have a color or be pointed. Thus, once again, this cannot be an instance of identity. The lightning must simply be composed of the electrical discharge. Finally we come to persons and their bodies. Philosophers have often thought that this was an identity, but we can demonstrate that it is not in the same way we demonstrated that the above are not identities. First, persons have attributes not shared by their bodies. A person may be in pain or know the president, but his body cannot be in pain or know the president. Second, the sortals "person" and "human body" have different criteria for reidentification. This is demonstrated by the fact that the body characteristically outlasts the person. Philosophers have traditionally wondered what the relation between a person and his body can be if it is not one of identity. What is it for a person to have a certain body? The answer is that it is for him to be composed of that body. That a person is composed of his body follows immediately from 4.4. A suitable φ might be "is a human body and encompasses all of the points of proprioceptive sensation of some person". This would seem to be an entirely adequate explanation of the intimate connection we feel between a person and his body. 4.3 Reidentification Objects falling under different sortals are in general reidentified in different ways. It follows that the account of reidentification given in section 3 cannot be applied in general to all physical things. We were led to talk about sortals to resolve one difficulty for that account, but this had led us to another difficulty, this time 8

Composition is reflexive—everything is composed of itself.

4. Sortals and Composition insurmountable. If the account in section 3 is correct for one sortal, it must be incorrect for most others. No single account of reidentification is adequate for cars, lumps of plastic, trees, and statues. In light of this we must reexamine our earlier account. We must ask what sort of thing is reidentified in the way described. What sortal is our account about? More or less following Strawson,9 let us say that basic particulars are those objects which can, in principle, be reidentified without first reidentifying anything other than the object under considera­ tion. To avoid an infinite regress, there must be basic particulars. If we ask how basic particulars are to be reidentified, all of the arguments of section 3 immediately become relevant. There must be some logical reason for reidentification which proceeds exclu­ sively in terms of attributes we can know the object to possess or not to possess without first reidentifying the object. The only such attributes are perceptual attributes, so there must be a perceptual prima facie reason for reidentifying basic particulars. This prima facie reason will clearly be just the reason described in section 3. Section 3 is an account of the reidentification of basic particulars. But what are basic particulars? It is easier to say what they are not. They are not trees, cars, statues, typewriters, etc. The criteria of reidentification for all of these sortals are more stringent than those for basic particulars. For example, consider cars. We cannot reidentify cars perceptually. Or more precisely, there is no per­ ceptual logical reason for reidentifying cars. This is because the attribute of being a car is not a perceptual attribute. It is quite true that we can often tell perceptually that a car we see at one time is the same car as one we saw at another time. But this is only because of certain contingent facts we have learned about cars. For example, suppose we witness a car coming toward us, passing us, and then disappearing into the distance. We can tell perceptually that we were seeing the same car all along. But this is only true because, as a matter of contingent fact, cars have characteristic appearances. We can generally tell whether something is a car by looking at it. This is relevant because a necessary condition for the car we see at the end of the period of observation to be the same as the one we began looking at is that what we were watching was a car all along. If at some point it ceased being a car, and then later became a car again, it could not be the same car. To get a 9Strawson

[1959], p. 38.

The Reidentification of Physical Things concrete example of this, suppose we are projected 100 years in the future and presented with fantastically shaped plastic cars quite unlike anything we have ever seen before. Because of our lack of familiarity with them, we may be unable to tell whether something is a car just by looking at it. Let us suppose further that in this future time there are factories that remanufacture cars. Old worn-out cars are placed on a conveyer belt which carries them first through a microwave oven where they are melted down into pliable lumps of plastic, and then the lumps of plastic are formed into new cars by passing through invisible force fields. The cars coming out of the factory are not the same ones as those that went in. The ones going in are destroyed, and the ones coming out are entirely new. Now suppose we are able to witness this entire process from some point above the conveyer belt. We may be quite unable to tell that what goes in ceases to be a car when it passes through the oven. To us, all the cars look like formless lumps of plastic. Hence we cannot tell perceptually whether what comes out is the same car as what went in. Therefore cars are not basic particulars. Next notice that although a different car comes out of the factory than went in, something has persisted throughout the remanufacturing process. There was an object which we could see and identify perceptually throughout the process. For example, if we just hap­ pened upon this factory without knowing that those plastic things were cars, we would observe a single object enter the factory, slowly have its shape changed, and emerge. It would never occur to us to deny that it was one and the same object throughout its so­ journ in the factory. This object was (i.e., constituted) at first a car, then ceased to be a car, and then became (i.e., came to consti­ tute) a new car. We watched one and the same object undergo all of these changes. What is this object which persists when the car does not? It is the lump of plastic. It is one and the same lump of plastic which first constituted the old car and then constituted the new car, and it is this lump of plastic that we reidentify perceptual­ ly. In general, what we might call lumps, hunks, pieces, or agglom­ erations of material are basic particulars.10 For want of a better 10 These are the only basic physical particulars, but they are not the only basic particulars. It will follow from the account of the concept of a person given in Chapter Nine that persons are also basic particulars.

4. Sortals and Composition term, I will call these perceptual objects, because they are objects that can be reidentified perceptually.11 The lump of material which is a basic particular is not to be identified with a quantity of material. If in the remanufacturing process the lump of plastic is converted into a lump of a different kind of plastic, it is still the same lump but not the same quantity of material. If even a single molecule is altered, we presumably have a different quantity of material, but not a different lump. In talking about basic particulars, the emphasis is on "lump", not "material". It is the lump itself that is reidentified perceptually, not the material. The relation between the lump and the material is simply one of composition—the lump is composed of the quantity of material. Although we are generally more interested in cars and statues than the basic particulars of which they are composed, the basic particulars are of preeminent conceptual importance. For example, on what basis do we ordinarily reidentify a car over an interval of continual observation? We judge perceptually that there is one and the same thing there all along, and we judge that it is (i.e., constitutes) a car throughout the interval of observation. This one and the same thing we judge to be there all along is a perceptual object. It is not the car, because it would still be there if at some point the car slowly melted down into a lump of metal. In general, the reidentification of nonbasic particulars is, in ways like this, parasitic on the reidentification of perceptual objects. Perceptual objects play an essential role in our thinking about other objects. This discussion of the reidentification of cars suggests a general logical reason for reidentifying nonperceptual objects. Where A is a sortal, it might seem that a conclusive reason for judging the A perceived at rt to be the same A as the one perceived at t2 is that the perceptual object of which the first A was composed is the same as the one of which the second A is composed, and that perceptual object constituted an A throughout the interval from to f2. This does seem to be correct for "car" and many other sortals,12 but it is not correct in general. There are sortals This is not meant to imply that only perceptual objects can be perceived. We can perceive the car just as readily as the lump of plastic. 12 I do not mean to imply that this is the only reason we can have for reidentifying a car. If it were, it would be impossible to disassemble a car and then regain the same car by reassembling it. 11

The Reidentification of Physical Things for which it fails. For example, consider "piece of sculpture". Suppose I am artistically naive and have never seen any abstract sculpture. I witness a man working a lump of clay into a weird shape. He then leaves it sitting on a pedestal. Along comes another man who reworks the shape in subtle ways. I do not realize that these are pieces of sculpture. As far as I know the two men are just idly manipulating the clay. If asked whether both men were working with the same object, I would certainly agree that they were. There is a perceptual object which persists throughout this process. In addition, there are the abstract sculptures. The first man created an abstract sculpture out of the clay. Thus the lump of clay, which is a perceptual object, came to constitute a piece of sculpture. When the second man came along and changed the shape, he thereby destroyed the first piece of sculpture and created a new one. Furthermore, his changes could have been such that at no point did the piece of clay cease to constitute a piece of sculpture. Every minute change left a piece of sculpture. But clearly the final piece was distinct from the first one. Thus, although the perceptual object continually constituted a piece of sculpture, the final piece cannot be identified with the original piece. Hence the sortal "piece of sculpture" does not have the same kind of criteria for reidentification as do sortals Uke "car". Different sortals function in different ways, and there is no reason why we should expect them all to be the same as regards reidentiiication. Although ^identification functions differently for different sortals, all sortals whose objects are "physical" are alike in one respect—their ^identification presupposes the reidentification of perceptual objects. This is because, by definition, to reidentify any nonperceptual object we cannot proceed simply on the basis of perceptual attributes. To reidentify that object we must know that something has some nonperceptual attribute. That something cannot be the object in question, because to know that it has the nonperceptual attribute would presuppose a prior ability to reidentify it, and we would have an infinite regress. But obviously the something cannot be an arbitrary object; it must be an object intimately connected with the object we are reidentifying. The only way to achieve this is to appeal to something which constitutes the object in question. For example, to reidentify a car we must at least know that the object we are considering is still a car. If we could only know this by reidentifying the car, we would

4. Sortals and Composition have an infinite regress. Instead we know it by knowing that the lump of plastic composing the car has a certain structure. We can know this without reidentifying the car. The plastic's having this structure is also a nonperceptual attribute, so to know it we must reidentify the lump of plastic. But this we can do without reidenti fying anything else, because the lump of plastic is a basic particular. In general, in reidentifying one object we may appeal to the attributes of another object of which it is composed; and to do that we may have to reidentify the second object by appealing to a third object of which it is composed. And so on. But this cannot go on indefinitely. At some point we must come to something we can reidentify without having to reidentify anything else—a basic particular. Otherwise we could never get started on the process of reidentifying the first object. This is why perceptual objects are conceptually necessary. Other physical sortals cannot stand on their own feet. Their reidentification must always lead us back ultimately to the reidentification of perceptual objects. Thus although the other sortals may be more interesting, perceptual objects are indispensable to their functioning. 4.4 Emergence and Inheritance We have seen a number of examples of composition, but there is an important difference between two kinds of composition. Sometimes it is a necessary truth that anything of one sort (anything falling under one sortal) must be composed of something of another sort. For example, it is a necessary truth that statues are composed of pieces of material—perceptual objects. Let us call such cases ones of analytic composition. Analytic composition can be defined precisely as composition in which the counterfactual biconditional of analysis 4.4 can be replaced by logical equivalence: (4.6) X is analytically composed OfYifi(Bzl) (35) (Bi) (3JJ): X = jxAix & Y = fxBix & φ(Υ) & [( 3 \χ)ΒιΧ & Cx) & prob(B/A & C) = 1 > prob(B/ A &^ C) & — ( 3 C*)[C*x-* Cx & Ο) (Λ*Γ D C*x) & prob (B/A ScC*) < piob(B/A & ^ C*)] & the way in which we know that (χ) (χεΓ D Cx) does not presuppose our first knowing that ( χ ) (χεΓ D B x ) " . Our fair sample defeaters are then propositions of the form ^(C, Γ, A, B).10 Our inductive evidence consists of observing that Γ consists of A's which are B's, i.e., Γ (A, B). Then the confirmation of (x) (Ax => Bx) by this evidence is defeated by a proposition of the form iJrCC, Γ, A, B). However, this is not the only way it can be defeated by attacking the fairness of the sample. Just as in the case of diminishers, if B is itself the conjunction of two predicates, B1 and B2, then if the sample is prejudiced in favor of just one of these predicates, this still defeats the confirmation. For example, suppose once more that we are examining wooden chairs, but now 101 would propose that this characterization of defeaters is also adequate for the case in which our sample consists of possible A's that are B's. It maysound metaphysically outrageous to talk about a set Γ of possible objects, but all of this can be regarded as a shorthand way of saying that the physically possible way of producing A's that are B's would result in their also being C's, etc. This is the way &(C, Γ, A, B) is to be understood in such a case.

Induction our conclusion is that they are both made of oak and have wicker seats. Once more, our sample comes from a factory producing only oak furniture. Then, although PTobiB 1ZA & C) = 1, we would not expect to have prob(#i & B2/A & C) — 1. Nevertheless, we would take the confirmation to be defeated. You cannot shore up a biased sample by adding unlikely things to your hypothesis. If the sample is prejudiced in favor of any part of the hypothesis, this defeats the confirmation. In general, defeaters which attack the fairness of the sample are propositions entailing &(C, r, A, B*), where B* is some predicate which is part of (i.e., entailed by) B. Let us define: (4.5) Dft2(F, Γ, C, B * , ( x ) ( A x = > B x ) ) iff { ( B -> B * ) & [P->*(C, TtAtB*)]). These type II defeaters, together with the type I defeaters already noted, exhaust the defeaters for inductive confirmation. Consequently, we can define direct confirmation by saying that P directly confirms (Λ:) (Ax => Bx) just in case P entails (possible) positive instances of (x) (Ax => Bx) and entails no defeaters. We must also require that (Ax => Bx) be "projectible". The meaning, and the reason, for this final requirement will be explained in the next section. (4.6) P directly c o n f i r m s ( x ) ( A x = > B x ) iff ( A x = > B x ) is projectible and ( 3 Γ) {[P —> T(A, 5)] & Dfti(Ρ, Γ, ( x ) ( A x ==> B x ) ) & — ( Ξ C ) ( Ξ B * ) D f t 2 ( P , Γ , C , B * , (x)(Ax^>Bx))}. Because of the particular nature of the defeaters involved in direct confirmation, it is possible to give a much simpler definition of instance confirmation than that provided by 4.1. In effect, clause 3 of 4.1 is rendered redundant by the character of type I defeaters: (4.7) P instantially confirms Q iff there are statements R i , . . . , R n directly confirmed b y P s u c h that ( R 1 & . . . & R n ) QP r o o f : If P instantially confirms Q , then the right side of 4 . 7 holds b y 4 . 1 . Conversely, if P directly confirms R , then P instantially confirms R. This is because if there were Slt ..., Sm supported by P as strongly as P supports R, and (S1 & . . . & Sm) ^ R, then we would have a type I defeater and P would not directly confirm R. And

4. Enumerative Induction by theorems 2.5 and 2.6, anything entailed by a conjunc­ tion of instantially confirmed statements will also be instantially confirmed. Thus, although we could not define the general concept of confir­ mation to be the deductive closure of elementary confirmation, we can define the more restrictive concept of instance confirmation as the deductive closure of direct confirmation. This completes our analysis of direct confirmation, and hence of instance confirmation. Finally, let us return to the paradox of the ravens. This paradox can be resolved by appealing to our results concerning fair sample defeaters. The paradox arises because observation of non-^4's that are non-B's directly confirms (x)(^ Bx ==> ^ Ax), and hence instantially confirms (Λ) (Ax D Bx). This becomes paradoxical when our sample consists, for example, of green plastic garbage cans, and we take that to confirm that all ravens are black. But, as Hempel observed, if we change examples this no longer seems so paradoxical. Consider "All good conductors of electricity are good conductors of heat". It is not in the least paradoxical to suppose we can confirm this by finding substances that are not good con­ ductors of heat and ascertaining that they are not good conductors of electricity either. Furthermore, if we shift the setting a bit it is not paradoxical to suppose we can confirm that all ravens are black by observing non-black non-ravens. Suppose we are some­ how provided with a catalog of everything in the universe listing certain of their important properties. K we go through the catalog picking out non-black things and then checking that they are not ravens, it certainly seems that we could confirm that all ravens are black. Then why is it paradoxical to suppose we can confirm that all ravens are black by going to a garbage-can factory and checking that the cans are all green as they come out? I think the answer is that observation of green garbage cans really does not confirm that all ravens are black. More precisely, although "c is a non-black non-raven" directly confirms "All non-black things are nonravens", "c is a green garbage can" does not, despite the fact that it entails "c is a non-black non-raven". The explanation is in terms of fair sample defeaters. Suppose our sample consists of green plastic garbage cans. The probability of being a non-raven given that something is a green plastic garbage can is 1. Thus our

Induction sample of non-black non-ravens is totally prejudiced against find­ ing any ravens in it. There was never any chance of finding any ravens in it, so we have a fair sample defeater. This is why, in­ tuitively, we would not regard this sample as confirming either that all ravens are black or that all non-black things are nonravens. Thus we resolve the paradox of the ravens. On the other hand, if we pick our sample of non-black things randomly from a catalog of everything in the universe, we will not automatically prejudice our sample against ravens, and for this reason we feel no reluctance about taking it to confirm that all ravens are black. Also, we are now in a position to explain the often-voiced intuition that the reason it is harder to confirm that all ravens are black by looking at non-black things than it is by looking at ravens is that the proportion of non-black things in the universe is so much greater than the proportion of ravens. Because there are so many more non-black things, unless we pick our sample very selectively (e.g., choose only birds), it is very likely that we will find everything in our sample to have some attribute C which en­ tails that it is not a raven. For example, picking things at random we are quite apt to end up with only inanimate objects, because most objects are inanimate, and a sample of non-black inanimate objects cannot confirm that all ravens are black. It seems that there are really two paradoxes of the ravens. The original paradox concerned how observation of non-black nonravens can confirm that all ravens are black. This paradox is re­ solved by appealing to fair sample defeaters. However, Hempel compounded the paradox by giving an incorrect explanation of it which then led him to maintain that observation of black nonravens also confirms that all ravens are black. The latter is simply wrong, and arises out of supposing that the generalizations that are confirmed inductively are properly symbolized with material conditionals.

5. A Theory of Proj ectibility 5.1 Nonprojectible Conditionals

I have argued that some form of the Nicod principle is the only serious contender for a correct principle of induction. But

5. A Theory of Projectibility now we come to what is the most serious difficulty for the Nicod principle. There are choices of the predicates A and B for which it clearly does not hold. Following Goodman [1955], let us say that the conditional (Ax => Bx) is projectible if it is directly con firmed by its positive instances. Contemporary philosophy owes a debt of gratitude to Goodman for bringing it irrevocably to our attention that there are conditionals which are not projectible. Goodman has pointed out at least three classes of nonprojectible conditionals:. 1. In general, conditionals having disjunctive antecedents are not projectible. For example, consider the conditional "If χ is either a mammal or a reptile then χ is warm blooded". If this conditional were projectible, then having observed a number of warm blooded mammals we would have observed a number of positive instances of it, and so would have confirmation for the generalization of the conditional. But the generalization entails that all reptiles are warm blooded, so that would also have been confirmed. But surely observation of warm blooded mammals gives us no reason at all to think that all reptiles are warm blooded. Thus the conditional "If χ is either a mammal or a reptile then χ is warm blooded" is not projectible. 2. Goodman showed how to construct a class of peculiar predi cates, which we can call "Goodmanesque" predicates, which in general cannot enter into projectible conditionals. Letting t be some particular time, e.g., 1:00 a.m., January 1, A.D. 2000, and given two predicates A and B, we can define a new predicate to mean χ is now A, and it is now earlier than t, or χ is now B, and it is not now earlier than t. Using this format, we can define the Goodmanesque predicates "grue" (green now and it is now earlier than t, or blue now and it is not now earlier than t), "bleen" (blue/green), "gred" (green/red), "emeruby" (emerald/ruby), "condulates" (con ducts electricity/does not conduct electricity). These predicates generate counterexamples to the Nicod principle. For example, consider the conditional "If χ is an emerald then Λ: is grue". Having observed many green emeralds, we have also observed

Induction many grue emeralds (given that it is now earlier than i, an emerald is grue iff it is green). If this confirmed the generaliza­ tion that all emeralds are grue, then it would confirm the con­ clusion that all emeralds after t will be blue. But the observation of green emeralds certainly does not confirm that conclusion. Thus these Goodmanesque predicates cannot enter into pro­ jectible conditionals, except perhaps in some extraordinary ways. 3. There are also some quite ordinary predicates that are ex­ cluded from projectible conditionals. Examples would be "has been observed" or "existed prior to It is certainly true that all the emeralds we have observed have been observed, but this in no way confirms the generalization that all the emeralds there are have been observed. Similarly, all the robins we have ob­ served have existed prior to the year 2000, but this gives us no reason to think there will be no new robins born after that time. These examples indicate that the Nicod principle cannot be maintained in its full generality. On the other hand, it seems clear that for some choices of A and B it can be maintained. Some con­ ditionals are projectible, and others are not. To get a correct ac­ count of inductive reasoning, we must give a characterization of those conditionals which are projectible, and then restate the Nicod principle so that it is restricted to those conditionals. This is why we defined direct confirmation as we did. The discovery of nonprojectible conditionals was a surprising one, but once they have been pointed out we can see that we should have expected there to be nonprojectible conditionals. In the introduction to this chapter, I argued that induction receives its validity from the fact that principles of induction are built into the justification conditions of our concepts. But there is no reason why induction should have to be built into the justification conditions of all our concepts. It should be quite possible to have concepts which cannot be handled inductively, and these are precisely the ones that generate nonprojectible conditionals. If this is correct, it ex­ plains very simply why there are nonprojectible conditionals. Goodman has shown that there are some nonprojectible con­ ditionals. At this point it still seems likely that most conditionals are projectible, with only a few pathological ones failing to be projectible. However, in what follows it will be shown that most conditionals fail to be projectible. Most, and perhaps all, con-

5. A Theory of Projectibility ditionals can be confirmed inductively, but only in ways more complex than countenanced by the Nicod principle. The projectible conditionals are those that receive their projectibility directly from the justification conditions of certain fundamental concepts into which the Nicod principle is built. Other conditionals can be dealt with inductively only insofar as this is a logical consequence of the projectibility of projectible conditionals. In order to defend the above account, I will divide the theory of projectibility into two parts. On any theory of projectibility there will be the basic projectible conditionals which receive their pro­ jectibility directly from whatever the ultimate source of projecti­ bility may be. Then by performing various logical operations on these basic projectible conditionals we can construct new projectible conditionals, and possibly other conditionals that are not pro­ jectible but can be dealt with inductively in more complex ways. Thus we can ask two questions: (1) Where do the basic projectible conditionals come from? (2) What logical operations allow us to construct new projectible conditionals from old ones, and by what means can we confirm nonprojectible conditionals that are built out of projectible ones? The answer to the second question con­ stitutes the logic of projectibility. It is convenient to discuss the logic of projectibility before discussing the source of basic projec­ tible conditionals, because this will give us some insight into the way projectibility functions. However, before we can do either we must construct a precise definition of projectibility. 5.2 The Definition of Projectibility Roughly, projectible conditionals are those whose universal generalizations are directly confirmed by their positive instances. However7 we cannot use "directly confirmed" in defining "pro­ jectible", because direct confirmation was defined in terms of projectibility. But we can use the more general notions of con­ firmation and elementary confirmation. A natural first attempt at defining "Α/Β" ("B is projectible with respect to A") would be: A / B iff (A a & B a ) elementarily confirms (χ ) ( A x => B x ) . However, this is not quite good enough. We want projectible con­ ditionals to be elementarily confirmed by evidence containing their positive instances just in case that evidence directly confirms them.

Induction This is what the notion of direct confirmation was intended to capture. The difficulty with the definition of projectibility proposed above is that (Aa & Ba) might elementarily confirm (χ) (Ax = > Bx) only "indirectly" in the sense that (Aa & Ba) entails a positive instance of some more general conditional that entails (x) (Ax => Bx). For example, let us define the Goodmanesque predicate "copronium" (copper/zirconium). Thentheconditional (metal(*) & coproniumO). ==> χ conducts electricity) should not be con­ sidered projectible. If we already know that some metal does not conduct electricity, then observing copper which now conducts electricity in no way confirms the conclusion that zirconium after the year 2000 will conduct electricity. On the other hand, if c is a piece of copper, then "metal(c) & copronium(c) & c conducts electricity" does confirm the generalization that all pieces of metal which are copronium conduct electricity. In other words, the latter conditional is confirmed by its positive instances. This is because "metal(c) & copronium(c) & c conducts electricity" entails "metal (c) & c conducts electricity", which confirms "All metal conducts electricity", and the latter entails "All pieces of metal which are copronium conduct electricity". Thus, although the latter conditional is confirmed by its positive instances, this confirmation is "inherited" from the confirmation of "All metal conducts elec­ tricity" and hence is defeated by finding a single piece of metal, say a piece of nickel, which does not conduct electricity. If "All pieces of metal which are copronium conduct electricity" were directly confirmed by its positive instances, its confirmation should not be defeated by finding a piece of metal that is not copronium which does not conduct electricity, because this is not an instance of any of the kinds of defeaters built into the concept of direct confirmation. Thus, if our definition of projectibility is to capture this intuitive notion, it must preclude the existence of such defeaters as this. This can be accomplished by simply listing the conditions under which evidence containing positive instances of (x) (Ax => Bx) may fail to confirm (λ) (AX —> Bx). This can happen only if the evidence contains either a type I or a type II defeater. Thus we can define projectibility as follows: (5.1) Α/Β iff for any statement P and set Γ, if P —> Γ(Α, B ) but P does not confirm (χ) (Αχ Ξ=> BX), then either Dft1CP, Γ, (Λ) (Ax => Bx) ) or ( 3 C) ( 3 B*) Dft2(P, Γ, B*, (χ) (Αχ ξ > BX) ).

5. A Theory of Projectibility 5.3

The Logic

of

Projectibility

Given our definition of projectibility, it becomes possible to establish a number of formal results. The major result will be that very few conditionals are projectible. However, it will also be shown that nonprojectible conditionals can be instantially confirmed, but not by using the Nicod principle. The way in which such a conditional can be confirmed will depend upon the conditional. 5.3.1 Closure conditions for projectibility. Now we can establish some simple results concerning how new projectible conditionals can be built out of conditionals we already know to be projectible. First, from theorem 2.5 and the definition of projectibility we have two equivalence principles. Letting symbolize logical equivalence: (5.2) (5.3)

Theorem: Theorem'.

Next we prove that projectibility is closed under the conjunction of consequents: (5.4) Theorem: If A/B and A/C then A/(B & C). Proof: Suppose A/B and A/C. Suppose and and 11

&

Cx)).

so . Suppose . Let S be the conjunction of type II diminishers entailed by P for the confirmation of by Then there are S . . . , Sn supported by P at least as strongly as supports and Then 1 }

As we are dealing with counterfactual conditionals, some of the logical moves made here and on the following pages may seem suspicious. For the rather strong counterfactuals symbolized by " s > " , the following principles seem to be true, and I assume them: (1) If and then if and , then if and , then if then 231 1 1

Induction

. [r(/l, B) & 5] supports Bx) as strongly as supports (x) (Ax by principle 4.3. Thus Dfti(P, r, which is contrary to supposition. Suppose that for some . Then and But then , so , which is contrary to supposition. Therefore, P confirms . Similarly, P confirms . Thus P confirms Cx)], and this conjunction is equivalent to . Consequently,

Goodman's observation that disjunctive antecedents create difficulties (5.5) gives Thereus:are A, B, C such that but not This gives us the rather surprising conclusion that projectibility is not closed under contraposition: (5.6) Theorem: There are D, E such that D/E but not Suppose otherwise. Choose A, B, C as in 5.5. Then and By 5.4, , and hence Finally, by theorems 5.2 and 5.3, (A v B)/C, which contradicts the choice of A, B, C, Theorem 5.6 indicates that we must be a bit careful what logical operations we perform on projectible conditionals in attempting to construct new projectible conditionals. Some operations work, e.g., we can conjoin consequents, but there are also some natural operations, such as contraposition, which do not work. Let us look at other natural operations. When we refute a generalization of the form by finding a counterexample, we often try to salvage things by either weakening the consequent or strengthening the antecedent. This suggests two possible principles: (1) (2) . Unfortunately, both of these principles Proof:

232

5. A Theory of Projectibility fail. Using Goodmanesque predicates, we can easily construct counterexamples. For example, we have already seen that although (metal(JC) => χ conducts electricity) is projectible, (metal(Λ:) & copronium(x). => χ conducts electricity) is not projectible. Thus proposal 1 fails. It is equally simple to construct counterexamples to proposal 2. It is natural to suggest that what went wrong with proposals 1 and 2 is that they were stated so generally as to allow the use of Goodmanesque predicates. Perhaps if they are stated in such a way that only projectible predicates are allowed, they will be true: (3) A/C&B/C. D (A&B)/C; (4) A/B&A/C. D A/(B\C). Proposal 3 may be correct. In fact, I conjecture that it is. However, proposal 4 fails. Astonishingly enough, projectibility is not closed under either the disjunction or the negation of consequents, nor is it under the formation of conditionals. First, consider disjunctions. Suppose A / B and A / C . Suppose we have examined lots of A's, found many of them to be either B's or C's, and not ascertained of any A's that they are neither B's nor C's. Would we then be justified in thinking that all A's are either B's or C's? It might seem so, but consider more carefully. Suppose the way we proceeded was first to examine lots of vl's to determine whether they were JB'S, and found that many were. But we did not go on to determine whether any of the A's that were not B's were C's. Analogously we examined lots of other A's and found that many were C's, but once again we did not go on to determine whether any of the A's that were not C's were B's. Under these circumstances, we have determined that many A's are either B's or C's, and have not determined that any 's are neither B's nor C's. But all we really know is that many A's are B's (and many are not), and many A's are C's (and many are not). There is no reason at all to think that the B's and C's exhaust the A's. To determine the latter, we must either examine many ^4's that are not B's and find that they are all C's, or examine many /i's that are not C's and find that they are all B's. If we did this, we would get the desired result. But this is because we apparently have (A & B)/C and (A & ^ C)/B and hence can determine inductively either that (;c) (Ax & ^ Bx. => Cx) or that (χ) (Ax & Cx. => Bx). Either of these conditionals entails the conditional (*) (Ax => .Bx ν Cx), so the latter can then be inferred logically. But the latter conditional cannot be confirmed directly by its posi-

Induction tive instances. We must always deal indirectly with conditionals involving disjunctions, deriving them logically from other con ditionals that are projectible. It follows that projectibility is not closed under the disjunction of consequents: (5.7) There are A , B , C such that A / B and A / C but not A / ( B

vC).

Nor is projectibility closed under the negation of consequents: (5.8) Theorem·. There are C / D such that C / D but not C / D. Proof: Suppose otherwise. Choose A, B, C as in 5.7. Then by hypothesis, A/ B and A/ -J C, so by theorem 5.4, A/(^ B & ^ C). By hypothesis again, Aj *-> (^ B & /— C), and hence by theorem 5.2, A / ( B ν C ) , which contradicts the choice of A , B, C . Theorem 5.8 is surprising. I would have supposed initially that whenever it is possible to establish inductively that all C's are D's, it is also possible to establish inductively that no C's are D's, but this is precisely what is denied by theorem 5.8. The explanation for this will be seen in the next section. Finally, we establish that projectibility is not closed under the formation of conditionals in the consequent: (5.9) Theorem·. There are C, D, E such that C/D, C/E, but not C / ( D D £ ) . Proof: Suppose otherwise. Choose A, B, C as in 5.7. Then by two applications of the hypothesis, A / ( B D C. D C). But (B D C. D C) is equivalent to (Β ν C), so A/{B ν C), which contradicts the choice of A , B, C . 5.3.2 Some conjectures on projectibility. We can pretty much sum up our discoveries regarding closure conditions for the conse quents of projectible conditionals in two words: Nothing works. The only truth function we have found the consequents to be closed under is conjunction. Furthermore, closure under conjunction is of little practical importance. Whenever closure under conjunction allows us to establish a conditional (χ) (Ax => .Bx & Cx) inductively, we could just as well establish directly the two conditionals (x) (Ax => Bx) and (x) (Ax ==> Cx), and then infer

5. A Theory of Projectibility O) ( A x ==> .Bx & C x ) logically. Thus closure of the consequents under conjunction in no way increases the class of propositions that can be established inductively beyond what we have using those basic projectible conditionals with which we begin. It seems likely that our basic projectible conditionals will be simple ones like (x is a raven => χ has eight toes). Unfortunately, the universal generalizations of these simple conditionals are at least mostly false. In order to obtain useful projectible conditionals we must have ways of weakening these basic ones. This will involve either strengthening the antecedent or weakening the con sequent. The most natural ways of weakening the consequent are by forming disjunctions or conditionals, but both of these moves are disallowed, by theorems 5.7 and 5.9. Let us turn instead to the possibility of strengthening the antecedent. The natural way to do this is by forming conjunctions. This takes us back to proposal 3, according to which if A/C and B/C then (A & B)/C. I will state this as a formal conjecture: (5.10) Conjecture: If A / C and B / C then ( A & B ) / C . This will give us more complicated conditionals like (Λ is a twofooted raven => χ has eight toes), which will at least have more chance of having true universal generalizations. It seems that we must also be allowed to introduce the negations of predicates into the antecedents of projectible conditionals. This suggests another principle: (5) If A/B then ^ A/B. However, if we assume con jecture 5.10, we can easily prove that proposal 5 is false: (5.11) Theorem'. If conjecture 5.10 is true, then there are D, E such that D / E but not ^ D / E . Proof: Suppose otherwise. Choose A , B , C as in theorem 5.5. T h e n b y hypothesis, ^ A / C and ρ and T * —> p, but T - ^ q and T * — > ^ q . If we suppose 7.1 to be projectible, then "O(p) & (T ρ) & ρ is true" confirms (p)[0(p) & (T p ) . D ρ i s t r u e ] . B u t " 0 ( p )" a n d " T —> p " are truths of logic. Consequently, "O(p) & (Γ p) & /? is true" is logically equivalent to p. Hence (by principle 2.7) ρ confirms ( p ) [ 0 ( p ) & ( T —» p ) . D ρ is true]. The latter entails [ 0 { q ) & (Γ q). D q is true], so this is also confirmed by p. But the antecedent of this conditional is a truth of logic, so the conditional is equivalent to q. Therefore, ρ confirms q.17 By applying the same argument to T * , we obtain the result that ρ also confirms ^ q . But ρ is a contingent statement, and as such cannot possibly con­ firm both q and q. If it did, it would confirm their conjunction (q & ^ q), and we have proven (theorem 2.8) that this is impossible. 18 Of course, I do not want to deny that, on a more reasonable construal of the term "theory", there are theories. I think that as the term is ordinarily used, there is no sharp distinction between theories and laws. Theories are just more general and less well confirmed. If a theory becomes sufficiently well confirmed, it becomes a law. What I am primarily concerned to deny in saying that there are no theories is that we need any principles of con­ firmation besides enumerative induction. if This itself seems manifestly absurd. If T is at all comprehensive, ρ and q may be statements about very different subject matters, ostensibly un­ related. Then ρ could not possibly confirm q.

Induction Consequently, generalization 7.1 cannot be projectible. But this means that if there were any theories they could not be confirmed because there would be no way to find out that their observational consequences are all true. Such theories would be worthless. Obviously, the actual theories of science are not like this. Real scientific theories are not "theories" in the sense of the received view. Is there any way to modify the received view of theories so that theories become confirmable? I take it to be essential to this view that theories are about theoretical entities which are in principle unobservable. So clearly enumerative induction will not work. We cannot confirm the hypotheses of a theory by enumerative induction, because we have no way (except by assuming the theory, which begs the question) of finding out that what the theory claims about theoretical entities is true in particular cases. The hypothetico-deductive method looks superficially like explanatory induction, so it might be supposed that we can confirm theories by explanatory induction. But that will not work either. In order to use explanatory induction we must know that the relations between theoretical entities hypothesized by the theory constitute a physically possible mechanism for producing the observable data. But if the entities discussed by the theory are truly unobservable, so that our only access to them is through the theory, then we have no way of knowing whether what the theory proposes about them is physically possible. On the other hand, if we drop the requirement in principle 3.4 (the principle of explanatory induction) that we know the physical possibility of the mechanism for producing the data, the resulting principle is equivalent to supposing that generalization 7.1 is projectible, so that is not defensible either. It must be concluded that there simply is no way to confirm theories insofar as they are about entities which are in principle unobservable. It seems inescapable that the real theories of science must not be about entities that are unobservable in the strong sense supposed by the received view. This conclusion can be defended on grounds completely independent of those adduced above. First we must make clear just what it is that the received view is affirming and we are denying. I think that the strong sense of "unobservable" embodied in the received view can be made precise as claiming that the only way to know of the existence or the states of a theoretical

7. Theories entity is by first confirming the theory. There is no way to know the states of such an entity prior to confirming the entire theory. Any weaker notion of "unobservable" would make it possible to confirm a theory by enumerative induction by ascertaining that what it asserts about the theoretical entities is true in particular cases. This construal of the observable/unobservable distinction requires it to be a sharp logical distinction. It has to do with the concepts of the entities, and not merely with contingent facts about them. It must be emphasized that "observable" (i.e., "not unob­ servable") is not the same thing as "perceivable". There may well be entities that we cannot perceive with our senses that are nevertheless observable because we can know of their states without having to confirm any theories by the hypotheticodeductive method. An example would be the molten core of the earth. We cannot see it, but we know a lot about it just by using enumerative induction and (derivatively) explanatory induction. There are at least two arguments to the effect that the actual entities of science are not unobservable in this strong sense. The first is a "slippery slope" argument, and the second is an appeal to actual scientific practice. Among the actual entities of science, there is no place to draw the line between observable and unob­ servable entities. It is undeniable that there is something different about the observability of tables and neutrinos, and one is tempted to record this difference by saying that the former are observable and the latter are not. But there is a continuum between tables and neutrinos. If a one-celled organism can only be seen through a microscope, is it a theoretical entity or an observable entity? What if some people can see it with the naked eye but others only with a microscope? For example, very nearsighted people can see hydras with the naked eye by holding a sample of swamp water very close to their eyes. Surely this is enough to make the hydra observable. But if all the nearsighted people in the world were suddenly to die, would the hydra become a theoretical entity? Certainly not if the observable/unobservable distinction is to be a logical distinction. Thus we must admit that entities seen only through microscopes can be observable. Then what about those entities which cannot be seen through optical microscopes but can be detected using an electron microscope? Surely the limits of resolution of optical

Induction microscopes are only contingent, and as such cannot contribute to a logical distinction between observable and unobservable entities. But even a single uranium atom can be observed through an electron microscope. By progressing further we can observe even smaller entities—electrons, mesons, etc.—using less "visual" in­ struments. As we progress down the scale of size, the entities do in some sense become less readily observable, but there is no point at which it makes any sense to draw a sharp line and say that those above the line are observable and those below the line theoretical. There is simply no basis for any sharp distinction between observ­ able and unobservable entities. Such a slippeiy-slope argument is not fatal to most distinctions, because most distinctions are not intended to be sharp. But it is essential to the observable/unobservable distinction that it be a sharp distinction of logical kind if it is to do the work required of it by the received view, and as such the slippery-slope argument is fatal. Of course, none of this is to deny that there are differences between different kinds of entities which might be marked by saying that some are observable and others unobservable; but these differences are only differences of degree and do not yield a sharp distinction, and more important, the differences are not such as to force upon us some entirely new principles of confirmation like the hypothetico-deductive method. I think it must be concluded that the observable/unobservable distinction, as traditionally construed, is an untenable dichotomy, and accordingly the hypothetico-deductive method is a philos­ ophers' fiction. But if we are not allowed to use the hypotheticodeductive method to confirm scientific theories, how are they to be confirmed? This only seems to be a problem when we accept the theoretical/observable dichotomy. We confirm scientific theories just as we confirm anything else—by using enumerative induction and (derivatively) explanatory induction. Once it is granted that there are no entities which are in principle unobservable, there is no reason why we cannot determine directly whether the principles of a theory hold in particular cases, and then by enumerative induction we can confirm the theory. For example, how was it discovered that negative electric charge comes in discrete bundles, subsequently dubbed "electrons"? Physicists began by confirming (by enumerative induction) various physical laws such as Cou­ lomb's law, laws governing the motion of a particle through a

7. Theories fluid, etc. These laws were confirmed by observation of macro­ scopic objects. Then Millikan performed his famous oil-drop experiment in which small negatively charged oil drops were subjected to an electrostatic field opposing the force of gravity. By measuring their rate of fall in a gas of known viscosity, Millikan was able to calculate the charges on the oil drops, and he confirmed (by enumerative induction) that all of the oil drops bore charges which were integral multiples of a certain small value. He thus confirmed inductively that negative electric charge came in discrete units. This is characteristic of the way in which physicists first acquired knowledge about atoms, molecules, and subatomic particles. They simply applied laws they had already confirmed inductively by appealing to larger objects, and in this way acquired knowledge about particular microscopic events. They then used this knowledge to confirm general laws about the microscopic world. These laws were subsequently used to acquire more particu­ lar bits of knowledge about the microscopic world, which in turn were used to confirm more general laws, and so on. For example, how would a contemporary physicist set about trying to establish a generalization regarding the production of muons under bombard­ ment of a target by energetic protons? He would perform experi­ ments which, in light of principles that have already been con­ firmed, would give him knowledge about particular instances of muon production. He would then use enumerative induction to confirm his general principle.18 I would urge that examination of the history of physics indicates that physicists have never used anything but enumerative induction (and explanatory induction) in establishing general principles about the microscopic world. There is no need for recourse to the hypothetico-deductive method or anything else more abstruse than enumerative induction to account for the reasoning that actually occurred in the history of physics. To support this contention in detail is an enormous task, far beyond the scope of this book, but I hope that the above brief discussion has at least made it plausible. 18 Notice

that this common sort of scientific practice would be com­ pletely mysterious to an instrumentalist. According to him, "statements" about theoretical entities are just tokens devoid of meaning (and hence not really statements). As such it would be in principle impossible to discover some new "law" about theoretical entities by enumerative induction, al­ though scientists do this all the time.

Induction Accordingly, one cannot defend the hypothetico-deductive method by saying that it is necessary to account for actual scientific reasoning. There is no good reason to think that scientific reasoning, or any other reasoning, requires any principles of induction other than those embodied in enumerative induction.

Chapter Nine

The Concept of a Person

I WITNESS a man hit by a truck. He is writhing about on the pavement and screaming. The splintered ends of broken bones are projecting through his torn flesh, and he is lying in a pool of blood. Being a philosopher, I ask myself, "How do I know that he is in pain?" This is representative of one type of problem regarding the concept of a person. The task of this chapter will be to analyze that concept. Pretty obviously, the concept of a person cannot be defined in terms of other purely physical concepts. It might be possible to analyze the concept of a person in terms of some mental concepts, but those mental concepts cannot themselves be defined in terms of physical concepts.1 This means that among our mental concepts there must be some ostensive concepts. Those concepts can be explained in terms of their justification conditions, and perhaps in no other way. Then it will be argued that, like the concept of a physical thing, the analysis of the concept of a person consists of giving an account of the justification conditions of propositions about persons.

1. Theories Regarding Our Knowledge of Other Minds We will begin the discussion of the concept of a person with one of the most venerable problems of philosophy: the problem of "other minds". In this section and the next I will consider how we can know that a person is in any particular mental state. For ex­ ample, how do we know that the man hit by the truck is in pain? 1 This, of course, presupposes the falsity of behaviorism, but that will be discussed in more detail in section 1.3.

The Concept of a Person More or less following Strawson,2 I shall say that an M-state ("M" for "material") is any state definable in the language of physics, and a P-state ("P" for "person") is any non-M-state of a person. This means that even such states as "going for a walk" or "trying to scale the North American face of El Capitan" are Pstates. A P-attribute is the attribute of being in a P-state, and a P-concept is the concept of a P-state. Not all P-states can reason­ ably be called "mental states", but the problem of how we can have knowledge of other minds is readily extended to the question of how we can know what P-states another person is in. Let us begin by surveying the traditional theories regarding this problem. 1.1 TheArgumentjromAnalogy Consider a statement ascribing a P-state to a person, e.g., "Jones is in pain". How do we determine whether such a statement is true? A traditional answer to this question is provided by the argument from analogy. According to this argument I discover inductively, on the basis of my own case, that in certain circum­ stances certain M-states tend to be accompanied by pain. When I subsequently witness a person in those circumstances manifesting those M-states, I have an inductive reason for thinking that he is in pain. There is a traditional response to this argument. It is claimed that it is a very weak induction because it is based upon a single instance—myself. I observe that something is true in one case, i.e., I observe that these M-states are accompanied by pain in myself, and I conclude that this is true of everyone. The standard reply to this objection is that it is not an induction based upon a single instance; rather, it is based upon all those cases where I have observed a person in those M-states and have been able to judge whether he was in pain. That those were all cases involving myself is irrelevant to the inductive argument. I think we must side with the defenders of the argument from analogy in this dispute. Their detractors seem to be assuming that the fact that all my inductive evidence concerns myself defeats the inductive argument. It could only do so if this were a reason for thinking my inductive evidence was not based on a fair sample. But the mere fact that the sample in question (my P-states) might not 2Strawson

[1959] talks about M-predicates and P-predicates. These are predicates ascribing M- and P-states respectively.

1. Knowledge of Other Minds be characteristic of others is not sufficient to constitute a defeater. If that were a defeater, all inductive arguments (regardless of subject matter) would be defeated analogously—it is always true that a sample might be uncharacteristic. What is required, to have a defeater here, is some concrete reason for thinking my P-states and M-states are not characteristic of people in general.3 In other words, the burden of proof is on the detractors to show that it makes a difference that all of the evidence concerns myself. They have not tried to do this, and certainly no general argument could be given to this effect. I think it must be concluded that this traditional attack on the argument from analogy involves a mis­ conception about induction. Let us turn now to a more sophisticated objection that has been made to the argument from analogy. This objection has been raised by Malcolm [1958], and he attributes it to Wittgenstein [1953]. We can recast the objection in our own terms as follows. Before we can have an inductive reason for believing a statement, that state­ ment must, of course, be meaningful. As I argued in Chapter One, this means that either we must be able to characterize the meaning of the statement in terms of its justification conditions, or else we must be able to state the truth conditions of the statement in terms of other statements whose meanings can be characterized in terms of their justification conditions. If the argument from analogy provides the only way to acquire knowledge of other minds, then it follows that we cannot give such an account of the truth con­ ditions of "Jones is in pain", so the meaning of that statement must be given by its justification conditions, i.e., by specifying what are logical reasons for believing it. In other words, in order for it to be possible to have an inductive reason for believing that a person is in pain, that statement must be meaningful, and so it must be possible to have a logical reason for believing that the person is in pain. Malcolm then reasons that if the argument from analogy were the only way of finding out that a person is in pain, then there would exist only inductive reasons for believing that; there would not exist logical reasons for the statement. But then the statement "Jones is in pain", having no logical reasons, would be meaningless, and hence the argument from analogy could not justify us in believing it either. Consequently, the argument from See section 4 of Chapter Eight for a more precise discussion of these "fair sample" defeaters. 3

The Concept of a Person analogy cannot explain how we can know that another person is in pain. It is only possible to have an inductive reason for believing that if it is also possible to have a more direct logical reason for believing it. Malcolm's argument contains an unstated presupposition. This is that the inductive reason itself cannot both be a logical reason for thinking that a person is in pain and be sufficient to determine the justification conditions of that statement. The phrase "the inductive reason" can be taken to denote either (1) the statement that Jones is behaving in a certain way under certain circumstances, or (2) the statement under clause 1 together with all of the evidence we have for thinking that people who behave in that way under those circumstances are in pain. Waiving for the moment any qualms about whether this is a good induction, we find that clause 2 is a logical reason, while clause 1 is a contingent reason. Let us ask whether clause 2 can exhaust the justification conditions of the statement "He is in pain". In general, it is clear that we could not have only an inductive reason for ascribing a particular concept to objects. In order to use enumerative induction, we must first ascertain in some other way that the concept tends to be correctly ascribable to things under certain circumstances. But of course, this condition is satisfied in the case of pain. We do not use an inductive argument to ascribe pain to ourselves, so we do not have only an inductive reason for ascribing pain to objects. It is just that we may have only an inductive reason for ascribing pain to certain objects, namely, to persons other than ourselves. Unless we can show that there is something wrong with this, Malcolm's argument cannot be made to work. I do not, at this point, see any way to demonstrate that there is anything wrong with this. However, there is another difficulty with the argument from analogy. The defenders of this argument are proposing that we can explain our knowledge of other minds without introducing any new epistemic principles having to do specifically with the mental. They are maintaining that our reasons for attributing P-states to others are derivative from general principles of induction which are universally valid, applying to all subject matter. But, as we saw in the last chapter, the principles of induction are not universally valid. There are many concepts, like "grue" and "bleen", which are not projectible, i.e., for which the principles of induction

1. Knowledge of Other Minds do not work. It was argued that insofar as induction is applicable to a concept, this is because the principles of induction are built into the meaning of the concept as part of its justification con­ ditions. When inductive reasoning works for different concepts, the inductive reasons are formally analogous in each case; but they are still different reasons which must be posited anew for each concept. There is no underlying rationale which simultaneously justifies all inductive reasoning. If induction is to be applicable to P-concepts> this must be because it is built directly into the justification conditions of certain P-concepts. Thus, even if the argument from analogy is correct, it does not succeed in explaining our knowledge of other minds—it merely describes that knowledge. To say that our knowledge of other minds is inductive is not to base it upon something we already have but to propose a partial analysis of P-concepts. Thus it becomes incumbent upon the de­ fender of the argument from analogy to show that P-concepts really are projectible and to show that it is possible for us to acquire all the knowledge we have of other minds by appealing to induc­ tion. This is not something we can take for granted. In other words, we need another argument to defend the argument from analogy. Rather than attempting to provide such an argument now, I suggest that it is best to examine the alternatives to the argument from analogy. 1.2 Scientific Realism

The argument from analogy attempts to give an inductive justification for our beliefs regarding the P-states of others. There is another type of inductive approach to this same problem. Some philosophers have argued that our beliefs about the P-states of others constitute a theory which is justified by the hypotheticodeductive method.4 I have already argued on general grounds that the hypothetico-deductive method is a philosophers' fiction and plays no true role in rationality. However, quite apart from that, there seems to be a devastating objection to the hypotheticodeductive method as a way of justifying our ascriptions of P-states to others. The problem is that if this were a correct account of our reasons for such ascriptions, it would make the P-states of others theoretical entities. On this account statements about the P-states of others would have no meaning independent of the theory—they 4 For

example, Castaneda [1966].

The Concept of a Person

would be given their meaning by the theory. But when we say that someone else is in pain, we mean to ascribe to him the same thing we ascribe to ourselves when we say that we are in pain. "Pain" means the same thing in both cases. When we learn to talk about pains, we do not have to learn two different meanings—one for ourselves and one for others. This implies that statements about the P-states of others cannot be given their meaning by the theory. Thus scientific realism cannot be correct. The only possible inductive argument of use in ascribing P-states to others is the argument from analogy. 1.3 Behaviorism It cannot be denied that behavior and other M-states and physical phenomena provide us with good reasons for ascribing P-states to other persons. That, after all, is the basis upon which we make such judgments. These reasons must be either contingent reasons or logical reasons. The position that they are contingent reasons is just the argument from analogy. Let us investigate the other possibility—that they are logical reasons. Because philos­ ophers have not traditionally recognized the existence of nonconclusive logical reasons, those who have supposed these reasons to be logical have generally supposed them to be conclusive. In other words, they have supposed that statements about the P-states of persons are logically entailed by statements about their various M-states. This is behaviorism. Behaviorists have generally gone further and maintained that statements about P-states are not just entailed by statements about M-states but are in fact logically equivalent to the latter. This is because they knew of no way to generate entailments except from truth conditions, and truth con­ ditions always yield equivalences rather than mere entailments. For this reason, behaviorists have traditionally maintained that it is possible to analyze the truth conditions of statements about P-states in terms of statements about M-states. However, I will con­ sider the weaker view which simply asserts that there are entail­ ments here, leaving open the question of whether they are gen­ erated by equivalences. To my mind, and I think it is generally agreed, the biggest problem with behaviorism comes from considering first-person ascriptions of mental states. For example, according to the behaviorist, there is some complex M-attribute P such that, if I have

1. Knowledge of Other Minds that attribute, it follows logically that I am in pain. But there can be no M-attribute the possession of which logically guarantees that I will feel a pain. What I feel and the way I behave are simply two different things. I can imagine behaving entirely differently than I do, perhaps consistently feigning pain to perfection when in fact I really feel nothing. No statement about my M-states can logically guarantee that I have any sensation at all. Consequently, my being in pain cannot be entailed by any statement about my M-attributes, and behaviorism fails. It is not my intention to claim that behavioristic analyses must always fail. There are some instances in which behavioristic analyses are much more plausible than in the case of pain. For example, it is not unreasonable to suppose that a behavioristic analysis of depression might be given. The most important part of being depressed seems to be how one acts, not how one feels. And even if it should turn out that all behavioristic analyses fail, I am quite certain that some statements about P-states at least entail statements about M-states. For example, "Jones is going for a walk" obviously entails something about Jones's M-states. But then it also follows that at least some statements about P-states are entailed by statements about M-states. This is because if P is a statement about P-states and it entails Q, which is a statement about M-states, then it follows that Q, which is also a statement about M-states, entails P, which is a statement about P-states. To this extent, then, the behaviorist would appear to be correct. Sometimes statements about M-states entail statements about Pstates. However, in order to give a complete account of our knowl­ edge of the P-states of others, he must maintain that statements about those P-states are always entailed by suitable statements about M-states, and in that he is mistaken. 1.4 The Criteriological Theory

If the argument from analogy fails, then statements attributing certain M-states to a person must constitute logical reasons for attributing corresponding P-states to them. The position that these logical reasons are conclusive reasons is behaviorism, but behavior­ ism fails. Let us consider instead the descriptivist position that we have nonconclusive logical reasons here. This position has re­ cently been held by a number of philosophers.® It has come to be 5

Malcolm 11958] held this, and he attributed the view to Wittgenstein

The Concept of a Person called the criteriological theory, because these philosophers sought what they called "logically adequate criteria". What they seem to have meant by this term is simply what I have called "prima facie reasons". According to the criteriological view, there are certain "basic" P-states such that, if X is one of them, then there corresponds a (possibly very complex) M-state Y such that a person's being in state Y is a prima facie reason for thinking he is in state X. Nonbasic P-states are regarded as being composites of these basic Pstates together with M-states. If one has rejected both the argument from analogy and behaviorism, this is the only possibility left— there must be such prima facie reasons. And this is indeed the basis upon which most adherents of the criteriological theory have defended their position. But the criteriological theory is subject to an immediate difficulty. Whether there are criteria for P-states or not, it is at least clear that we do not ordinarily employ them when making selfascriptions of P-states. I do not have to look at my own behavior to know whether I am in pain. Thus first-person and third-person ascriptions are made on entirely different bases. One cannot help but wonder how they get hooked together to make up the justification conditions for a single concept. This hooking together cannot be simply a matter of convention. We cannot construct a new P-concept by taking the mode of self-ascription from one (say, being in pain) but the third-person criterion from another (say, being happy). Such a conglomerate would not be a P-concept. If I ascribe a concept to myself because I feel pain, and to another because he acts happy, there is no reason at all to think that there is a P-state that we share. Thus we cannot construct P-concepts in such a haphazard way. There must be some connection between the means of self-ascription and the means of other-ascription. If we ask what the connection must be, the answer seems fairly clear: a P-concept must be "coherent" in the sense that, were we to use the means of third-person ascription on ourselves, we would get approximately the same result as we do with our actual means of first-person ascription. The only apparent way of finding out that a concept is coherent in this sense is by induction. Thus it seems [1953]. Strawson [1959] argues for such a view, and it was endorsed by Shoemaker [1963].

1. Knowledge of Other Minds that before we can even know whether we have a P-concept we must be able to discover inductively what M-states are correlated with the P-state in our own case (thus testing a proposed criterion for coherence with the means of self-ascription). But now the criterion drops out of the picture as irrelevant. We have the right criterion just in case it is what we would arrive at inductively, and in that case we do not need it because we could get the same result proceeding purely inductively. Thus the criteriologist seems to be forced into endorsing the argument from analogy. A criteriologist might be tempted to agree that we have to proceed inductively in tying together a means of self-ascription and a criterion, but maintain that I have described things backward because we begin with the criterion rather than the means of selfascription, and then proceed inductively to tie the means of self-ascription to the criterion rather than the other way around. But this only sounds different; it is not really different. In order to proceed Inductively in establishing a correlation between two classes of things,, we must be able to observe both classes of things independently, Thus, whether we start with the criterion or firstperson observations οϊ the P-State i we must be able to reidentify the P-state when it recurs in our own experience before we can relate it inductively to the Criterionl and -3S long as that is possible, the criterion is unnecessary. At this point a defender of the criteriological theory is apt to respond that we are not free to just make up new P-concepts at will. We are somehow constrained to use (only?) those already entrenched in our language as it now exists.® These come with the means of third-person ascription already built into them, because we are taught them in terms of those means of third-person ascription. For example, consider how we teach a child the concept of pain. When we judge that he is in pain, we say "You are in pain", and when he volunteers that he is in pain when we judge that he is not, we tell him that he is not in pain. In order to do this we must be able to tell when he is in pain, which, of course, we do in terms of some means of third-person ascription. We teach the concept of pain in terms of that means of third-person ascription, and consequently it cannot help but be built into the concept. The above response only helps if it is supposed that it is not eThis

response has been made to me by John Turk Saunders in corre­ spondence on this topic.

The Concept of a Person incumbent upon us to check the coherence of those P-concepts we are taught by others as part of learning our language. The only reason I can see why that might be so is if coherence is, after all, at least partly a matter of convention. We saw that it cannot be entirely a matter of convention, but it might be proposed that it is a matter of convention for certain basic P-concepts (those en­ trenched in our language), and then the coherence of proposed new P-concepts can be tested against these basic concepts. For example, the reason we cannot build a P-concept by combining the means of first-person ascription of pain and the means of thirdperson ascription of happiness is that we already have two different concepts in our language which employ those means of ascription. If we did not, presumably, it would be open to us to construct the new concept described. So the basic assumption that is required to make the criteriological theory work is that the coherence of Pconcepts is, at least ultimately, a matter of convention. We are only precluded from introducing a new P-concept in a certain way when doing so would conflict with our conventions regarding other Pconcepts that we already use. Without this assumption, the criteriologist is inexorably driven in the direction of the argument from analogy. The above position is plausible, but a little reflection indicates that it is mistaken. It is implied that there is no objective test of the coherence of a new P-concept other than its not conflicting with P-concepts we already employ. But this is obviously wrong when we consider the case of a person who acquires a new P-concept from his own experience, without it being taught to him by other people. Anyone who has been troubled by some peculiar recurring sensation and, perhaps fearing an impending heart attack or some other physical ailment, has attempted to talk to his doctor about it is in precisely the situation of acquiring the concept of a P-state from his own experience and looking about for corresponding Mstates with the help of which to explain to his doctor what P-state he is in. Obviously, no M-states which might ultimately be found to be associated with the P-state can bear any logical connection with the concept of that state. I expect that some philosophers will profess to find this example unintelligible, and will ask how one could ever know that it was the same sensation that was recurring on several occasions if he were unable to relate it to any M-states. I can only reply that having been in this sort of situation myself, I

1. Knowledge of Other Minds can assure the reader that there is not the slightest difficulty in reidentifying such a sensation when it recurs, even though one has not yet found any M-states which tend to be associated with it. Two lessons are to be learned from this example. First, given that a person in the above situation can and does proceed in­ ductively in looking for means of third-person ascription of his P-state, it follows that there is an objective standard against which the coherence of a means of third-person ascription for this con­ cept is to be tested, i.e., the inductive standard. A criteriologist might reply that this is only possible against the background of a basic set of P-concepts already entrenched in our language,7 but in the proceedings just described, no other P-concepts were used at any point. It does not seem to make any difference whether we have any other P-concepts. Thus I think it is unreasonable to main­ tain that there is no objective standard of coherence for means of first-person and third-person ascriptions. Given this, there seems to be no way to defend the criteriological theory against the objection that it presupposes the validity of the argument from analogy. The second lesson to be learned is that no particular M-states need be built in as criteria in P-concepts that are acquired from our own experience, and hence the criteriological theory will not work for such concepts. But it is surely a contingent fact that we are taught any particular P-concept rather than acquiring it from our own experience, so it follows that the criteriological theory is mistaken in general. The upshot of this is that there need be no logical connection (prima facie or conclusive) between a P-concept and any Mconcepts. This does not mean that there are no P-concepts that bear any logical relationships to M-concepts. We have already seen examples of some that do, e.g., "is going for a walk". The conclusion is rather that there exist P-concepts that are not logically related to M-concepts. For the above reasons, it seems to be true in general that those states which we could reasonably call "phe­ nomenological states" are not logically connected with any Mstates.8 On the other hand, if we turn to those P-states that are not phenomenological states, we find all sorts of connections between 7This

is maintained in Saunders and Henze [1967], pp. 188-191. An attempt will be made in section 3 to say more precisely what a phenomenological state is and how P-states are related to phenomenological states. 8

The Concept of a Person them and M-states, and I think it is this that made the criteriological view initially plausible. It must be concluded that the criteriological theory, as it is normally understood, does not give an adequate account of our knowledge of the P-states of others. There may be some P-states for which we can find the postulated prima facie reasons, but they do not exist for the vast majority of P-states. Nevertheless, I do not believe that the criteriological theory is completely wrong. I think that it errs only in the level of specificity of the prima facie reasons it postulates. Bleeding, grimacing, and crying out cannot be part of a prima facie reason for judging that a person is in pain, because it is only a contingent fact that persons in such M-states are usually in pain. There is a simple reason why this must be the case. The M-states which the criteriologist wants to employ as prima facie reasons are states of one's body. But it is only a contingent fact that persons have bodies at all like they do in fact have. There would be no logical absurdity in supposing persons to have bodies that are incapable of bleeding, grimacing, or crying out, or in general incapable of being in any particular M-state we may care to specify. If, as the criteriologist supposes, these prima facie reasons constitute the ultimate content of our P-concepts, then there would be no way to ascribe these concepts to persons whose bodies are very different from our own or to persons who lack bodies altogether. Is it a necessary truth then that persons have bodies like our own, or for that matter that they have bodies at all? Are ghosts a logical absurdity? I think not. Suppose that whenever people die the room is filled with a ghostly chill. This chill can be roughly located in terms of the spatial region in which it is felt, and it tends to move about. Furthermore, in the presence of this chill, one is apt to hear a voice, apparently emanating from empty space. It is possible to carry on conversations with this voice, and the voice is capable of informing us of facts regarding things that have happened both in the presence of the chill and in the presence of the person who died. The voice seems to evidence anger, pleasure, etc., associated with the goals and desires of the person who died, and in all respects behaves just as if it were the voice of that person. Can there be any doubt that we would judge that person to have become a ghost, a disembodied person? Furthermore, if we lived in a world like this, and recognized the presence of ghosts, it would also become possible for us to

1. Knowledge of Other Minds discover the existence of ghosts that had never been embodied. This indicates that it is not only logically possible for a person to become disembodied, it is also logically possible for a person to have never had a body. To take another example, suppose a world in which, rather than having corporeal bodies, persons have associated with them multi­ colored magnetic fields. These magnetic fields have definite shapes, are affected in predictable ways by their surroundings, and in turn are able to affect their surroundings in predictable ways. These persons would not be unlike we are, except that they would be acted upon and would in turn act upon their environment in terms of magnetic phenomena rather than mechanically. Although nonmaterial, these magnetic men would still have bodies of a sort— their magnetic fields—but the M-states of these bodies would be quite different from our own M-states. The important thing about these examples is that it is only a contingent fact that we aren't like these incorporeal beings. We could just as well have been ghosts or magnetic men as human beings. And this would not hinder us in our ability to judge one another's. P-states. For example, we could judge that a ghost is enraged either because he tells us that he is or because he begins making vases fall off shelves, lamps tip over, windows break, etc. This indicates once more that the way in which we make these judgments cannot be by appealing to particular M-states which constitute prima facie reasons for judgments about P-states. It is contingent what M-attributes we are even capable of possessing. Our magnetic men could not bleed. Although they might do something which, once we knew how to interpret it, we could call a "grimace", their grimaces need bear no physical resemblance to our own. This indicates that the concept of a grimace is not really an M-concept. The concept of a grimace is defined not in terms of its physical characteristics but rather in terms of the P-states to which it is generally a response. We might well discover that the grimaces of our magnetic men look more like our own smiles. Analogously, our magnetic men may cry out, but what constitutes ciying out for them may be physically completely unlike our own crying out. They may communicate in terms of radio waves rather than sound waves, so that crying out for them would consist of emitting certain kinds of bursts of radio waves. We could only discover this by seeing what P-states tended to elicit such a re-

The Concept of a Person sponse. This indicates that crying out is not an M-state either, but like grimacing it is a P-state. It is the P-state which consists of being in whatever M-state is generally elicited by certain other P-states (e.g., being in pain) and proceeds in terms of whatever channel of communication plays the role for them that vocal communication does for us. The above indicates that although grimacing and crying out might be part of a logical reason for ascribing pain to others, they are not M-states. In order to make use of such a logical reason, we would first have to find out what M-states constitute instances of grimacing and crying out, and that we could only do inductively. The simple criteriological view, according to which there are always specific M-states which constitute prima facie reasons for ascribing P-states to persons, will not work. If there is a physical prima facie reason for ascribing these P-states to others, it cannot involve any specific M-states. We discover what M-states are associated with what P-states. But per­ haps therein lies a prima facie reason. The prima facie reason is that the various M-states which have been discovered to be con­ nected with the P-state are possessed by the person in question: (1.1) "S possesses an M-state which tends to be accompanied by being in the P-state X" is a prima facie reason for me to think that 5 is in state X. The above seems to be the only way to make the criteriological theory work. More specific prima facie reasons than these cannot exist. But now, how do we discover what M-states are associated with a P-state? We must make this discovery before we can ascribe P-states to others, so the only possible answer is that we begin with our own case. We discover that in our own case certain M-states. tend to be accompanied by the P=State in question, and then we can use this, together with our prima facie reason to judge when others are in that state. Of course, this is generally made easy for us by the fact that we are usually taught the concept of the P-state in terms of the M-states that are connected with it, and so we learn both the concept of the P-state and what M-states are associated with it simultaneously. This seems to be the only way to make the criteriological theory work, but it is an unfashionable view. Perhaps most contemporary philosophers would deny that one can begin from one's own case,

1. Knowledge of Other Minds and they give very sophisticated arguments to support this. For example, Strawson [1959], pp. 99-100, has argued that you cannot have a P-concept until you know how to ascribe it to others. Con­ sequently, you cannot begin from your own case. I find Strawson's argument extremely opaque. There is a perfectly good sense in which the criteriological theory just elaborated does not violate Strawson's principle. If you have a P-concept, then you do know how to ascribe it to others—by seeing what M-states are associated with it in your own case and judging that others are in the P-state when they are in the corresponding M-states. I am sure that Strawson would disallow this, but then I do not know how to interpret his principle so as to make it defensible. The best way to cut through all of these arguments and see that one can begin from his own case is to find a clear example of this, and we do not have to look far to find such examples. Most P-concepts are taught to us, and when that happens there is an important sense in which we do not begin from our own case—we are taught the corresponding Mstates right along with the P-concept. But sometimes we do acquire new P-concepts that have not been taught to us. Recall the man who was troubled by the recurring sensation that he wanted to tell his doctor about. That example indicates that it is quite possible to begin from one's own case in determining what M-states are as­ sociated with a P-state. All of the arguments which purport to disprove this must simply be wrong. If we agree that the argument from analogy must be rejected, I think it must be concluded that there are physical prima facie reasons for ascribing P-states to others. After all, it is on the basis of their M-states that we do make such ascriptions, and our judgments are certainly defeasible. But where the "traditional" criteriological theory errs is in supposing that these prima facie reasons make reference to any specific M-states. Rather, the prima facie reason consists of the person being in any M-states which we have found are generally accompanied by the P-states in question. For example, philosophers have bandied about the term "pain behavior", supposing that to be some physically specifiable be­ havior such that when a person behaves in that way his so behav­ ing constitutes a prima facie reason for thinking he is in pain. On the contrary, a person's manifesting pain behavior is a prima facie reason for thinking he is in pain, but we cannot specify a priori what constitutes pain behavior. Pain behavior is whatever behavior

The Concept of a Person tends to be accompanied by pain, and we have to discover what kind of behavior that is.

2. Our Knowledge of the P-States of Others Now I will try to bring some order to the above observations and provide a systematic account of our knowledge of the P-states of others. 2.1 The Equivalence of the Argument from Analogy and , the Criteriological Theory I have argued that there are only two possible theories regard­ ing our knowledge of the P-states of others—the argument from analogy, and a watered-down version of the criteriological theory. But this does not yet solve the problem. We would like to know which of these two theories is the correct one. I shall now argue that both are correct, because they are equivalent. The apparent differences between them are chimerical. We have here two theories regarding the justification conditions of statements about the P-states of others. The defenders of the criteriological theory give a fairly concrete account of those justi­ fication conditions. I have argued that the most defensible version of the criteriological theory is that proposing principle 1.1 (al­ though, as we will see shortly, that principle must be modified a bit). On the other hand, the argument from analogy proposes merely that those justification conditions are derivative from the principles of induction. But what does this amount to? It amounts to saying that a person's being in the M-states picked out by prin­ ciple 1.1 constitutes an inductive reason for thinking he is in the corresponding P-state. In other words, we get as inductive reasons just the prima facie reasons proposed by principle 1.1. But, of course, inductive reasons are prima facie reasons, so the argument from analogy yields precisely the prima facie reasons that are forth­ coming from the criteriological theory. Now the only apparent difference between the argument from analogy and the criteriological theory is that the former labels these prima facie reasons as inductive reasons, while the latter maintains that they are directly constitutive of the P-concepts themselves, giving, as they do, the justification conditions of those concepts. But what does this difference really amount to? The traditional

2. Knowledge of the P-States of Others defender of the argument from analogy would have denied that these reasons are constitutive of the P-concepts, because he thought of induction as consisting of universally valid principles standing above all concepts. But as we have seen, that is an erroneous picture of induction. Not all concepts are projectible. When a concept is projectible, this constitutes part of its justification conditions, and hence is partially constitutive of that concept. Consequently, a defender of the argument from analogy ought to regard these reasons as directly constitutive of the P-concepts. Once again, we find no dispute between the criteriologist and the defender of the argument from analogy. In conclusion, I do not see any way to distinguish between the argument from analogy and the criteriological theory. Let us briefly rehearse the argument. The argument from analogy is the position that our reasons for judgments about the P-states of others are contingent reasons. The only alternative is to maintain that those reasons are logical. They cannot be conclusive reasons, so they must be only prima facie. Furthermore, these prima facie reasons cannot make reference to any specific physical attributes. What physical attributes are involved must be discovered inductively. Thus, a sufficiently general criteriological theory resolves itself into the original argument from analogy, and that becomes the only possible theory of our knowledge of the P-states of others. 2.2 Persons and Bodies We have been discussing how we can know what P-states a person is in. This question presupposes the ability to identify something as a person, or to know that there is a person present. Now let us turn to the more general question of how we can know that there is a person present who is in a certain P-state. A person is not the same thing as his body. When a person dies, he ceases to exist, but his body may continue to exist unchanged. And it will be argued in section 4 that it is logically possible for a person to change bodies. So we cannot say of a particular body that it is a person. Let us talk instead of bodies being inhabited.9 How do we tell whether a particular body is inhabited? Clearly, we do so by seeing whether it behaves like an inhabited body. If it 9 It was argued in Chapter Six that the relation between a person and his body wherein he inhabits the body is one of composition—the person is composed of his body.

The Concept of a Person exhibits pain behavior, acts angry, happy, bewildered, etc., under the appropriate circumstances, then we judge that the body is in­ habited. Judging whether it behaves in these ways is a matter of judging, on the basis of principle 1.1, that if the body were in­ habited, then its inhabitant would be in pain, angry, happy, bewildered, etc. We distinguish between a person and his body. We seem to make an analogous distinction between a dog and its body, and even a lobster and its body. For any living thing, it is possible for it to die, and hence cease to exist, without its body ceasing to exist. Thus the living thing must be distinguished from its body. What makes something a living thing is that it has P-states. For example, lobsters at least feel pain. How do we tell when a lobster feels pain? He behaves in a manner not unlike we behave in situations analo­ gous to those that would cause us pain. In other words, we use the same prima facie reasons to ascribe P-states to nonhuman animals as we do to persons. This indicates that we should reformulate principle 1.1 as a principle about bodies and living things, and not just about persons: (2.1) "Body B possesses an M-state which tends to be accom­ panied by a living thing being in the P-state X" is a prima facie reason for me to judge that body B is inhabited by a living thing in state X . A person is simply a living thing that exhibits sufficiently many and sufficiently complex P-states. A person must be capable of believing, thinking, desiring, contemplating, admiring, etc. I doubt whether any precise list can be made of the P-states that are necessary to make something a person—the concept is not that precise—but this should be sufficient to characterize this rather fuzzy concept. Now another problem arises. Principle 2.1 can only be employed in connection with persons who have bodies, but it has already been argued that it is logically possible for persons to be disem­ bodied. Ghosts are not a logical absurdity. But as long as they exhibit other physical manifestations, we would not find it too difficult to judge the P-states of disembodied persons. We would make these judgments on the same basis as our judgments regard­ ing embodied persons—by seeing that they exhibited physical

2. Knowledge of the P-States of Others manifestations which we had found to be accompanied by the P-states in question. This indicates that we must generalize principle 2.1 so that it can be employed in connection with disembodied persons: (2.2) "The present state of affairs is of a kind that I have found to be accompanied by the presence of a living thing in the P-state X" is a prima facie reason for me to think that there is present a living thing in state X. 2.3 The Problem of Copersonality Principle 2.2 is not yet enough to account for all of our judgments concerning the P-states of others. By using principle 2.2 we can judge that there is a person present who is in some particular P-state. But we often want to judge of one and the same person that he is in several P-states. For example, using principle 2.2, we can judge that there is a person present who is in pain, and we can judge that there is a person present who is angry. How can we tell whether this is one and the same person? A simple answer might be that we use principle 2.2 in connection with the complex P-state consisting of being both angry and in pain. But this use of principle 2.2 would require us to have previously observed people in that complex state, which is unnecessary. It is only necessary that we have observed some in pain, and others angry. We can then put together what we have learned to tell that a person is both angry and in pain. How do we do this? This is the problem of copersonality—how do we determine that two P-states are states of the same person?10 Some philosophers have taken the problem of copersonality seriously. But they have generally thought that there was a problem for both the first-person and third-person judgments of copersonality, and have sought the solution in some phenomenologically discoverable relation of copersonality. Other philosophers have 10There are actually two problems of copersonality: what makes two simultaneous P-states states of the same person, and what makes the Pstates of a person at one time copersonal with his P-states of a later time? We might call these "horizontal" and "vertical" copersonality, respectively. In this section I am only concerned with horizontal copersonality. Vertical copersonality becomes a special case of the problem of reidentifying persons and will be dealt with in section 4.

The Concept of a Person denied that there is a problem of copersonality,11 on the grounds that we can only identify a P-state in terms of the person whose state it is, and hence to know of the existence of a P-state is automatically to know whose state it is. As we shall see, both of these views are in error. The latter philosophers are right that there is no problem in the first-person case, but there remains a difficulty in the third-person case. Let us begin by considering first-person judgments of coper­ sonality. Bertrand Russell [1914] found our ability to make such judgments mysterious, and in need of explanation. He sought such an explanation in the content of those P-states themselves, main­ taining that there was a relation of "being experienced together" that we are aware of in our P-states, and that it is this relation which makes one's P-states all states of oneself. As Shoemaker points out, this leads quickly to absurdity. On this theory, if I were to feel a pain, but not observe it to stand in the relation of copersonality to my other P-states, I would have to conclude that it is not my pain. But if I feel it, then of course it is my pain.12 In the first-person case, there simply is no problem. If I judge of each of two P-states that I am in them, then it follows automatically that they are copersonal. There is nothing further that I have to establish to know this. It is tempting to suppose that the same thing is true of thirdperson judgments of copersonality. It may seem that we can only identify a P-state in terms of the person whose state it is. There is no way to refer to a P-state without making reference to the person who possesses it. Consequently, it seems that if we know of the existence of a particular P-state, we automatically know whose it is, and so the problem of copersonality cannot arise. Un­ fortunately, this is a mistake. Using principle 2.2, we may know of the existence of a person in a certain P-state, e.g., in pain. We know this on the basis of the person's body. In one sense, we do know who it is that is in pain—the person whose pain it is. This is a uniquely referring description. But of course, this is not a very informative answer. And we do not automatically know of another way to refer to the person in question. Whether that person is Jones is a substantive question which we have not yet seen how to settle. Analogously, we may know that there is someone who is angry. Then the question of whether the pain and anger are 11 For

example, Ayer [1964],

12 Shoemaker

[1963], p. 104.

2. Knowledge of the P-States of Others copersonal is simply the question of whether the person who is in pain is the same person as the one who is angry. How can we know whether this is so? It may seem that there is a simple solution to this problem. If, using principle 2.2, we judge the presence of a person in pain on the basis of the M-attributes of a body, we automatically know of another term referring to the person in question, viz., "the person whose body this is". Then if we judge, on the basis of that same body, that there is a person who is angry, we know that the person who is angry is also the person whose body this is. Thus it follows that the pain and the anger are copersonal. This is certainly correct as an account of our actual judgments of copersonality. We at­ tribute two P-states to the same person iff it is the same human body that is involved in our coming to know of both P-states. If this is a logical reason for judgments of copersonality, then it seems that our problem is solved. However, there are difficulties with the suggestion that coper­ sonality is simply a matter of sameness of body. First, persons need not have bodies, but we could still make judgments of coperson­ ality. It would not be particularly difficult to tell that a ghost is both embarrassed and angry. Second, even if we restrict our attention to embodied persons, our formula for copersonality will not work for two reasons: (1) it is a contingent fact what constitutes a body, and the ability to determine this presupposes the prior ability to make judgments of copersonality; (2) it is not a necessary truth that two persons cannot share a single body, in which case same­ ness of body would not be a guarantee of copersonality. Taking these points in order, first consider what it is for a certain physical object to constitute a person's body. A person's body is, roughly, the locus of those M-states which are regarded as manifestations of his P-states. If it were a necessary truth that a person's body must consist of a single physical object, this locus might not be difficult to find, but there is no such necessary truth. For example, we can imagine persons whose arms have wings which allow them to fly off and perform tasks remote from the rest of the body and then come back to roost on the shoulders. In general, there is no reason why a person's body might not consist of an extensive array of disconnected "parts" working in unison. To discover the extent of this array, we must discover, e.g., the extent to which pains exhibited by the different parts are copersonal. Thus judgments of

The Concept of a Person copersonality cannot in turn presuppose judgments of sameness of body. Going on to the second point, the human brain consists of two hemispheres joined together. To a large extent, the right hemi­ sphere controls the left side of the body, and the left hemisphere controls the right side of the body. As a cure for severe epilepsy the two hemispheres are sometimes split apart. The results have been carefully studied, and are fascinating. The two hemispheres seem to communicate with one another via external clues. For example, the right hemisphere hears with the left ear what the left hemisphere is saying.13 Now, suppose we have a child who is born with a split brain. Each hemisphere is dormant, asleep, for half of the child's waking day, and during that time the other hemisphere controls the child's entire body. Because different things happen during their distinct periods of control, the two hemispheres would be conditioned differently. They would exhibit different knowledge, different memories, different desires and as­ pirations, etc. When the child reaches maturity, a change occurs and the two hemispheres no longer sleep while not in control. Each still controls the body for half of the body's waking day, but each is able to report what went on while the other hemisphere was in control. Furthermore, because the two personalities are so dif­ ferent, a marked antagonism develops between the two hemi­ spheres. The one hemisphere, who happens to have become a neurophysiologist, begins experiments designed to "drive that meddlesome intruder from my body". The other, who is an opera singer, goes to the police and asks them to prevent this happening. Is it not extremely plausible that here we have two persons in a single body? So far, this indicates that it is logically possible for two persons to inhabit a single body. However, if this is to be an obstacle to judgments of copersonality, it must be shown that the single body can simultaneously exhibit the P-states of both persons. An ad­ dition to the above story yields the desired result. The neuro­ physiologist continues his experiments with the object of ultimately gaining permanent control of his body, and one day announces that he believes he has succeeded. The test will come when his twelve hours of control are up and the opera singer appears. But -3 See Gozzaniga [1967] for a fascinating nontechnical discussion of some of these experiments.

2. Knowledge of the P-States of Others the scientist is only partly successful. Different parts of the body seem to struggle against one another. The opera singer wishes to go to a party, so the body starts to walk into the bedroom to dress. But although the legs cooperate, the arms grasp wildly at furniture to drag the body back. The carefully modulated tones of the opera singer protest that something terrible is happening and that he cannot control his arms. But the voice breaks off in midsentence and the raspy voice of the scientist announces with glee that he is gaining control of the body. Then one of the arms suddenly begins cooperating with the legs and seeks to pry the other hand loose from the davenport to which it stubbornly clings. Would we not conclude that this body is inhabited by two persons each striving for dominance? Accordingly, although the body might exhibit both pain and anger, there is no guarantee that it is one and the same person who is both angry and in pain. Thus copersonality cannot be defined in terms of sameness of body. Of course, we do base our judgments of copersonality on same­ ness of body. But we must be proceeding inductively. We discover that our own P-states are related in certain ways to a certain physical object—our body. We observe other similar objects around us which have sufficiently similar attributes to allow us to employ principle 2.2 and judge the presence of other persons. Then we discover that there is only one person associated with each of these bodies, and hence we can base subsequent judgments of copersonality on sameness of body. But how do we discover that there is only one person associated with each body? The simplest way is to observe that this is true of our own body and then conclude inductively that it is true of others as well. But there must be other evidence that is relevant too—if this were the only relevant evidence it would be impossible for us to ever discover cases like the split-brain example in which sameness of body is not a guarantee of copersonality. To see what else is relevant, consider the scientist and the opera singer once more. If we know both well, we might not have too much trouble sorting out the different mental states exhibited by their common body and assigning each to the appropriate person. If the opera singer is a very meek but emotional person, and the body suddenly bursts into tears of despair, we would attribute the despair to the opera singer. If the scientist is renowned for his fiery temper, and the body begins throwing things about and bellowing with rage,

The Concept of a Person we would attribute the rage to the scientist. If a curious mouse happens upon the scene, and the body jumps screaming upon an end table, we would attribute the fear to the opera singer (who is known to be afraid of mice). These attributions are based upon the "naturalness" of certain temporal sequences of P- and M-states. If a person has in the past exhibited both meekness and emotionality, we would not be surprised to see him burst into tears of despair in a difficult and threatening situation. A person who in the past has exhibited a tendency to become angry easily does not surprise us when he again becomes angry. If a person has previously acquired knowledge of certain facts, we expect him to act accordingly. K he has previously exhibited certain likes and dislikes, we expect his future behavior to be consonant with those likes and dislikes. And so on. What constitutes "acting accordingly" is something we learn inductively, starting ultimately from our own case. We learn that certain sequences of P- and M-states are common, and others, e.g., laughing when stuck with a pin, are extremely unlikely. It is on the basis of the "naturalness" of the sequence of P-states that we can determine, at least to some extent, which P-states to attribute to the scientist and which to the opera singer, even though both concurrently inhabit the same body. The explanation for this is that (1) we have a strong inductive reason for thinking that copersonal states can be placed in "natural" sequences, and (2) we have a strong inductive reason for thinking that those Pstates exhibited by a single body which can be placed in a natural sequence are copersonal. The latter is the basis upon which we group the states we attribute to the scientist and the states we attribute to the opera singer as we do. The former is the basis upon which we deny that they can all be grouped together as states of one person. My conclusion is that our third-person judgments of copersonality are always inductive. This is the only way to explain the variety of situations that can arise regarding copersonality. 2.4 How P-Concepts Are Possible I have given a rudimentary account of how we acquire knowledge of the P-states of others. But it may be felt that a difficulty arises for this account. Consider my ascription of pain to myself and to another person. I know that I am in pain because

2. Knowledge of the P-States of Others I feel the pain, but I cannot feel another person's pain. I ascribe pain to him on the basis of his behavior. And both of these bases for ascribing pain seem to arise directly from the concept of pain. There is some temptation to suppose that in the first-person case I am simply reporting my feeling, but in the third-person case I am actually reporting the other person's behavior. But this cannot be right, because there is always the logical possibility that the other person is pretending. Whatever his behavior, it is not logically sufficient to guarantee that he is in pain. We want to say that it is not his behavior we are reporting but what he feels; but how can this be possible on the basis of his behavior? How can we be ascribing the same thing to ourselves and to others when we do so on such different bases? Described in this way, P-concepts seem mysterious. How can there be one concept that is ascribed to different subjects in such totally different ways? However, the air of mystery disappears once it is realized that the line between the two ways of ascribing P-concepts is improperly drawn as the distinction between our­ selves and others. We can, and do, ascribe P-concepts to ourselves in precisely the same way we ascribe them to others. For example, upon witnessing a home movie taken of myself several years ago, I may well judge on the basis of my behavior that first I was angry and then amused. My grounds for this judgment would be just the same as my grounds for making this judgment about someone else. In other words, the line between the two ways of ascribing P-states does not coincide with the distinction between my P-states and the P-states of others, but rather with the distinction between P-states I am currently experiencing and other P-states. The former cate­ gory contains only P-states of myself, but the latter category contains P-states of both myself and others. Once we have seen the proper place to draw the line, the two ways of ascribing P-concepts should no longer seem perplexing. P-concepts become completely analogous to other familiar con­ cepts. For example, consider the concept of a red object. There are two ways of ascribing this concept to an object: (1) by seeing the object and judging its color on the basis of the way it looks to us; or (2) by appealing to various inductive reasons which we have acquired ultimately by using the first reason. Analogously, there are two ways of ascribing a P-concept to persons: (1) by ex­ periencing the P-state and making a judgment on that basis; or

The Concept of a"Person (2) by appealing to various inductive reasons which we have acquired ultimately by using the first reason. For either concept, our second kind of reason arises simply out of the fact that the concept is projectible. In the case of P-concepts, it has nothing special to do with other persons. It is, in effect, just a logical coincidence that this is the only way of ascribing P-concepts to other persons, and should be viewed as no more significant for the nature of the concept than the fact that we can only employ inductive reasons in ascribing colors to objects we do not see. The distinction between the two ways of ascribing P-concepts is just the distinction we should expect from the fact that they are projectible coupled with the fact that, in the relevant sense, we can only experience our own P-states."

3. The Analysis of P-Concepts 3.1 Projeciibility In the above sections we constructed elaborate principles like 2.1 and 2.2 that were intended to formulate the actual inductive reasons which the defender of the argument from analogy or the criteriologist would propose. However, these principles are both too weak and unnecessary. First, they are too weak because there are ascriptions of P-states to others which do not proceed in terms of these principles. Once we can make some judgments about the P-states of others on the basis of principle 2.2, it then becomes possible to make other judgments about their P-states on a different basis altogether. Recall once more the example of the man who experienced a strange sensation in his chest but was unable to find any M-state associated with it. This makes principle 2.2 inapplicable, but does not guarantee that others will be unable to discover what sensation he has. Although he may not be able to find any M-states such that whenever he is in those states he has the sensation, he may nevertheless be able to describe the sensation quite well. He might be able to tell us, "It is a fluttery sensation I get in my chest, accompanied by a burning sensation in my throat. When I get it I feel light-headed, and get a panicky feeling as if I cannot breathe properly. When I lie down it goes away." 14My

274

thinking on this was made much clearer by reading Ayer [1953].

3. The Analysis of P-Concepts Given a fairly detailed description of this sort, another person might respond, "I know exactly what you mean; I get the same sensation whenever I eat my wife's cooking." What is happening here is that through the use of principle 2.2 we learn what he means by the various elements of his report. Thus we can understand what attributes his sensation has even though we do not know what sensation it is. It is then possible for us to discover inductively that only one kind of sensation has all of those attributes, and hence we can identify his sensation on the basis of its attributes. This identification of a P-state in terms of its attributes is some­ thing not covered by principle 2.2, so it seems that that principle should be augmented by another. However, this is unnecessary. We can replace any such principle, and also principle 2.2 at the same time, by a simple postulate of projectibility. What these principles do is attempt to formulate various aspects of our inductive reason­ ing about P-states, but all such principles can be eliminated in favor of a single principle saying that P-concepts are projectible. Such a principle of projectibility automatically licenses the reason­ ing carried out in accordance with these other principles. However, slight reflection indicates that a postulate of pro­ jectibility must be restricted—not all P-concepts can be projectible. We can construct Goodmanesque P-concepts, and the strictures elaborated in the last chapter against disjunctions, conditionals, etc., apply just as much to P-concepts as to any other concepts. I shall argue that, among P-concepts, it is basically the concepts of phenomenological states that are projectible.15 (Of course, the notion of a phenomenological state must be made more precise than it has so far been made.) What gives any P-concept its characteristic "mental flavor" is that it makes reference somehow to phenomenological states, generally mixing them with M-states. In general, P-states are hybrids of phenomenological states and M-states. Accordingly, inductive reasoning with respect to Pconcepts other than those of phenomenological states is derivative from the projectibility of phenomenological concepts and physical concepts. To defend this, I will examine phenomenological states more carefully. What is desired is a complete account of the justification conditions of statements ascribing phenomenological states to per15 Although not exclusively—for example, we have already seen that the concepts of perceiving and remembering are projectible.

The Concept of a Person sons. This should make it clear that phenomenological concepts are projectible. 3.2 Phenomenological States Phenomenological states (at least as I shall use the term) are those P-states that one can pick out introspectively. To use a phrase of Quine's, they are those that are "tinged with awareness". Phenomenological states are those that are subject to introspective demonstrative reference. In other words, they are those states to which the argument in Chapter Four is applicable. Accordingly, the external judgment "I am in that phenomenological state" is in­ corrigible. The simple incorrigible external judgment that one is in a par­ ticular phenomenological state does not get us very far. We must compare states, reidentify them, label them (e.g., "This is a state consisting of my being appeared to redly"), identify them with the states of others, and correlate them inductively with M-states and other P-states. Let us consider how such judgments are possible. 3.2.1 Identifying phenomenological states with one another. We often identify two phenomenological states as being the same. For example, I may judge that the pain I now feel is just like the pain I felt yesterday. This sameness is not the sameness of the individual states themselves (i.e., we are not saying that one and the same state has recurred), but is rather sameness of type. This becomes clear when we notice that we can judge two simulta­ neously existing phenomenological states to be the same. For ex­ ample, I may judge that I am simultaneously presented with the same color in two different parts of my visual field. This is the same notion of sameness as is employed in reidentifying a present state with an earlier one. Let us call this "phenomenological sameness". Phenomenological sameness does not require that two states have all their attributes in common—otherwise it would be nu­ merical identity rather than sameness of type. What it requires is that they be "phenomenologically indistinguishable". If we could lay them side by side, we would find no phenomenological differ­ ence between them. In other words, they have all their phenom­ enological attributes in common where phenomenological attributes

3. The Analysis of P-Concepts

are those attributes that are, in some sense, "directly discriminable". We have no difficulty deciding which attributes are relevant to whether two states are the same. For example, time of occurrence is not relevant, but, e.g., if the states involve visual objects, then the colors of the presented objects are relevant. However, we need a more precise characterization of phenomenological attributes. This is not difficult to come by. Phenomenological attributes are those which are relevant in judgments of phenomenological sameness, so they must refer to features of phenomenological states to which we have the same kind of direct access as we do to phenomenological states themselves, i.e., to those features which are subject to introspective demonstrative reference. This immediately gives us the following: (3.1) An attribute A of phenomenological states is a phenomenological attribute iff our being in a state having attribute A is itself a phenomenological state. Thus we can think of the phenomenological attributes of a state as referring to other simpler phenomenological states which are "logical parts" of the state in question. Furthermore, if we wish to judge that two phenomenological states are the same in certain phenomenological respects, although not entirely the same, this amounts to judging that certain states which are logical parts of the larger states are phenomenologically the same. For example, we may judge that two perceptual states are the same with respect to the presented object being red, although not in other respects. This amounts to saying that both perceptual states contain as parts of them states of being appeared to redly, where these latter states are phenomenologically the same. Given two simultaneously existing phenomenological states, how is it possible for us to know that they are the same? This is a very perplexing question. The sameness of two states seems so manifest that it is hard to imagine anything more basic to which we might appeal. Nevertheless, as we shall see, there is something more basic. We have a definition of phenomenological sameness in terms of phenomenological attributes: (3.2) The phenomenological states X and Y are phenomenologically the same iff they have the same phenomenological attributes.

The Concept of a Person Consequently, to ascertain whether two states are the same, we need merely check their phenomenological attributes. Unfortu­ nately, this does not take us as far as we might suppose. The difficulty is that definition 3.2 is essentially circular. We defined the notion of a phenomenological attribute in terms of phenomeno­ logical states, in such a way that to say two states X and Y have all of their phenomenological attributes in common is merely to say that every phenomenological state which is logically part of X is phenomenologically the same as some state which is logically part of Y, and vice versa. Accordingly, we have defined phenom­ enological sameness in terms of phenomenological sameness. This is not to say that 3.2 is useless, but only that it is incomplete. Definition 3.2 still achieves something, because it relates the phe­ nomenological sameness of composite states to the phenomenologi­ cal sameness of their parts. But we still need some way of judging the sameness of those "simple" states which do not have other phenomenological states as parts. Being appeared to redly is an example of such a simple state. In the case of simple phenomenological states, it seems there can be nothing more manifest than their distinguishability or indistinguishability. In other words, our ability or failure to distin­ guish between two simple phenomenological states in which we find ourselves is itself a phenomenological state. For example, upon being presented with two colors, our being able to distinguish between them is something that stands out clearly as an object of introspective awareness. So, too, is our not being able to distinguish between the colors. So these two states of being able to distinguish or not being able to distinguish are phenomenological states. And it seems clear that it is these states which provide the basis for our judgments of phenomenological sameness. If we cannot distinguish between two phenomenological states, we judge them to be the same, and if we can distinguish between them, we judge them to be different. The latter is clearly a conclusive reason: (3.3) "X and Y are simple phenomenological states and I can distinguish between them" is a conclusive reason for me to think that X and Y are not phenomenologically the same. Somewhat surprisingly, our inability to distinguish between two simple states is not a conclusive reason for judging them to be the

3. The Analysis of P-Concepts same. It is well known that although two colors, sounds, pains, etc., may be indistinguishable by direct comparison, they may be distinguishable by comparison with a third object. For example, although two colors may seem exactly the same upon direct com­ parison, we may be able to find a third color which is indistin­ guishable from one of the original colors but distinguishable from the other. This is just the psychological phenomenon of "just noticeable differences". Clearly such "indirect" distinguishability is sufficient to guarantee that two indistinguishable phenomenological states are not phenomenologically the same. Hence indistinguishability cannot guarantee sameness. On the other hand, indistinguishability is certainly the basis upon which we judge sameness, so it must be a prima facie reason: (3.4)

and Y are simple phenomenological states and I can­ not distinguish between them" is a prima facie reason for me to think that X and Y are phenomenologically the same.

"X

The only defeaters for this prima facie reason seem to arise out of the above sort of indirect distinguishability. If X and Y are phenomenologically the same, then they ought to be distinguish­ able from precisely the same states: (3.5) "There is a state Z which is distinguishable from one of X and Y but not the other" is a conclusive reason for thinking that X and Y are not phenomenologically the same. By using this reason, we can discover inductively that two states are not the same because, although we are not now in such a state Z, we could be. For example, I may know that the colors of two adjacent slabs that look the same to me are different because, although there is no third color present with the help of which to distinguish between them, nevertheless it is possible for such a third color to be present. I may know of this possibility either directly, by having compared the slabs with such a third color on other occasions, or I may have some general inductive reason for this judgment based perhaps on a measurement of the wave lengths of the light reflected by the slabs. The above principles seem to be correct, but at first glance they do not appear to explain how it is possible for us to know

The Concept of a Person that two states X and Y are the same. The problem is that, in order to have this reason for judging them to be the same, we must know that we cannot distinguish between them. Our being unable to distinguish between them is itself a phenomenological state, so to have this reason is to know that we are in a certain phenomenological state. But to know that one is in a certain phenomenological state would seem to involve the ability to identify the present state as being the same as other states of the same type, and then we are off on the slippery slope of an infinite regress. Fortunately, there is a mistake here. I have asserted that my reason for thinking that X and Y are the same is my knowing that I am in the state of being unable to distinguish between them. But this claim is ambiguous in a way that most claims about persons' epistemic states are ambiguous—it does not make clear whether my knowing that I am in that state is external or internal knowledge, i.e., whether I must know of the state in which I find myself that it is the state properly called "being unable to distinguish between X and Y" or whether I must simply find myself in a state which is in fact properly so called.16 A bit of reflection indicates that only the latter is required. In judging that two states are the same because I cannot distinguish between them, clearly it is irrelevant whether: (I)I know what words describe my being in the state of being unable to distinguish between them; or (2) I recall any other cases of being in the state of being unable to distinguish between two states; or anything else of this sort. All that is important is that I am aware of being in the state which is in fact the state of being unable to distinguish, and that awareness is external knowledge. As we saw in Chapter Four, such external knowledge is incorrigible. Hence we have succeeded in anchoring our judgments of phenomenological sameness in an incorrigible foundation, and thereby have explained how our knowledge of phenomenological sameness is possible. We can encapsulate this by expanding principle 3.4 as follows: (3.6) The external belief that I am in the simple phenomenological states X and Y and am in the state of being unable to distinguish between them is a prima facie reason 16For a discussion of external and internal knowledge and belief, see Chapter Four.

3. The Analysis of P-Concepts for thinking that X and Y are phenomenologically the same. 3.2.2 Reidentifying phenomenological states. So far I have talked about identifying two present states as being phenomenoIogicaIiy the same. We also make such judgments about states separated in time. In other words, we can reidentify one state as being the same as an earlier state.17 What is our basis for such reidentification? Sometimes we recall a previous state, in the sense of almost literally "reliving" it, and then compare the resulting "image" with the present state. For example, in deciding whether two separated surfaces look the same color to me, I may carefully examine one, try to hold an image of it in my mind, and then turn to the other surface and compare my image with the color the other surface looks to me. The image I have is equally an image of my being presented with the colored surface, and insofar as I am comparing not just the color but also the color's being per­ ceptually presented rather than, e.g., imagined, I am comparing the remembered phenomenological state with the present one via principle 3.6. This aspect of the reidentification proceeds just like the identification of two contemporary states. When we turn to the memory that is involved in the reidentification, we see that it is propositional memory. What we remember is that our earlier phenomenological state was the way our present image represents it. Thus we need no new principles to account for this aspect of our reidentification. It proceeds entirely in terms of the prima facie reasons enumerated in Chapter Seven. The above sort of reidentification wherein we have a memory image of the earlier state is to some extent unusual. It is perhaps more common to simply recognize a present state as being the same as an earlier state. For example, upon feeling a peculiar sensation or seeing an unusual color, I may simply recognize it as being the same as a sensation I had earlier or a color I saw once before. This recognition is possible without my being able to resurrect any kind of image of the earlier state prior to my being in the present state. I may be quite unable to recall the previous state "in a vacuum". In this case the reidentification proceeds entirely in terms of memory, without having to appeal even to principle 3.6. 17 This

is reidentification of the state type, not the individual state.

Tfae Concept of a Person Thus once again we need no new principles to account for reidentification. The above cases of reidentification are cases in which we actually remember the earlier state rather than merely remembering that it was a state of a certain sort. But we can also reidentify on the latter basis. For example, I may recall that on a certain occasion I heard a clear tone which was C-sharp. When on a later occasion I again hear a C-sharp tone, I can judge that my state of being presented with that tone is phenomenologically the same as my earlier state, even though I do not remember the earlier state. Or to take another example, I may discover inductively that whenever I eat onions I get a characteristic queasy feeling. Then upon eating onions once more I may know inductively that my present feeling is the same as what I had last time I ate onions, even though I don't remember how I felt then. These ^identifications are essentially inductive. In learning to use the label "C-sharp", I must discover inductively that all tones so labeled are phenomenologically the same. Thus upon encountering two such tones, I can conclude on the basis of their both being C-sharp that they are phenomeno­ logically the same. Similarly, in the second example I discover inductively that the queasy feelings I get on any two occasions of eating onions are phenomenologically the same. Thus it is possible to reidentify phenomenological states inductively as well as directly via principle 3.6 and prepositional memory. 3.3 The Projectibility of Phenomenological Sameness The fact that we can reidentify phenomenological states in­ ductively is of extreme importance. It requires that the concept of phenomenological sameness be projectible, and that is all we need for the account of knowledge of other minds given in sections 1 and 2. It was argued there that our basis for ascribing P-states to others is inductive. I have maintained that most P-states are hy­ brids of M-states and phenomenological states, so the problem of dealing inductively with P-states is ultimately the problem of dealing inductively with phenomenological states. On this account, our basis for ascribing a phenomenological state to another person is our discovering that whenever one is in a certain M-state he is in that phenomenological state. To discover inductively that when­ ever one is in a certain M-state he is also in a certain (type of) phenomenological state is to discover that, on any two occasions

4. Personal Identity of persons being in that M-state, the persons will also be in phenomenological states which are phenomenologically the same. All we need to make this sort of reasoning possible is for phenom­ enological sameness to be projectible, and we have just seen that it must be. Thus there is no obstacle to the sort of inductive reason­ ing about other minds that has been proposed by the defender of the argument from analogy and the modified criteriologist. Of course, this only secures inductive reasoning for phenomeno­ logical states—not for P-states in general. In order to deal induc­ tively with other P-states we must first see how they are to be analyzed in terms of phenomenological state and M-states, and then deal with them as we do in general with pseudoprojectible conditionals.18 Although some of these analyses may be difficult to come by, there would seem to be no difficulty in principle here. I think we can regard the problem of other minds as solved.

4. PersonalIdentity Now we come to one of the most perplexing problems regarding the concept of a person. What has come to be called "the problem of personal identity" is the problem of reidentifying persons. There have been basically two theories regarding personal identity. On the one hand there have been those philosophers who maintain that questions of personal identity are to be settled by an appeal to bodily identity. On the other hand there have been those philos­ ophers who have sought "mentalistic" criteria of personal identity. Among the latter philosophers, the predominant candidate for a mentalistic criterion has been some form of memory criterion. In the ensuing discussion I shall come down rather weakly on the side of the memory theorists. 4.1 First-Person Reidentification The problem of reidentification can be fruitfully divided into two problems: how do we reidentify other persons, and how do we reidentify ourselves? Let us begin with the latter question. How can I know that I am the person picked out by some term that makes reference to an earlier time? Sometimes I reidentify myself in the same way as I reidentify others. For example, looking at a 18 For

a discussion of pseudoprojectibility, see Chapter Eight.

The Concept of a Person newspaper picture of a crowd I may see my own face. How do I know that I am the person of whom that picture was taken? In this case I reidentify my body on the basis of appearance and then reidentify myself on the basis of the reidentification of my body. Here my reidentification is substantially the same as what I would do in reidentifying another person. But philosophers have usually wanted to maintain that we have another more direct way of re­ identifying ourselves. We don't generally have to examine our body to know that we are the person who performed some action in the past, or witnessed some past event. We have some "internal" way of knowing this. Thus we are led to memory. Locke and other modern philosophers tried to make personal identity consist of the possession of a continuous stream of con­ sciousness. As they recognized, the only way to make sense of this notion was in terms of memory of phenomenological states. To say that a stream of consciousness is continuous is to say that at each instant one can remember some of his past phenomenological states. In attempting to make this both a necessary and sufficient condition for personal identity, these philosophers encountered insurmountable difficulties. For example, a person suffering from amnesia may remember none of his earlier phenomenological states, but he does not thereby lose his personal identity. More simply, upon awakening from sleep I may remember none of my earlier phenomenological states. Consequently, this cannot be a necessary condition for personal identity. Is it at least a sufficient condition? Not as stated. It is a mistake to suppose we cannot remember the phenomenological states of others. If I know that another person feels anger or humiliation, I can subsequently remember his anger or humiliation. But this is not truly a counter­ example to what these philosophers had in mind. When they talk about remembering an earlier phenomenological state, they really mean remembering their being in that state. This is a sufficient condition for identity. If I remember that I was in a certain phenomenological state, then of course it follows that I was the person who was in that state. However, this is not terribly helpful for most cases of reidentification. Characteristically we are more interested in knowing whether, e.g., I am the person who broke the window yesterday. I do not generally come to know this by knowing anything about the phenomenological states of the person who broke the window. Rather, as Shoemaker [1963] and others

4. Personal Identity

have called to our attention, I may simply remember that I broke the window. Memory gives us much broader access to our past than merely to our past phenomenological states. The way in which memory helps us to settle questions of per­ sonal identity is not entirely straightforward. Although I may re­ member that I broke the window, I do not thereby remember that I was the person who broke the window. In order to conclude the latter, I must also know that only one person broke the window. I may know this on the basis of memory, too, but that is unneces­ sary. This is simply an impersonal fact about the past, and I can know it in any of the ways that I acquire historical knowledge. In general, the way memory helps me to settle questions about my personal identity is by providing me with facts about my past which show that I satisfied a certain description which I know, on some grounds or other, was only satisfied by one person. What I am proposing here is not really a memory criterion of identity at all (except perhaps in a very attenuated sense). Rather, I am propos­ ing that memory can give us facts about our past history, and then we can use these facts to infer that we uniquely satisfied certain descriptions.19 There is a common response to the above observations. It is replied that it is a trivial tautology that if I remember doing some­ thing then I did it, but that this is epistemologically irrelevant to questions of personal identity. A necessary condition of remem­ bering something is that it be true, and so it seems that before I can know I really remember something (rather than merely re­ calling it) I must first know that what I purport to remember is true, and in order to know that I must first settle the question of personal identity that is involved in the memory claim.20 Common though this argument is, it should be obvious what is wrong with it. It overlooks the existence of prima facie reasons. We have al­ ready seen that my recalling-that-P is a prima facie reason for me to believe-that-P, and for me to believe that I remember-that-P. If it were not, historical knowledge would be impossible. In order to know that I remember, I do not first have to establish the truth of what I claim to remember; my having the recollection auto19 In effect, I am agreeing with Shoemaker [1963] that we do not employ any criterion in reidentifying ourselves. If there is a criterion for reidentification, it is only for third-person reidentification. 20 For example, this argument is given by Williams [1956].

The Concept of a Person matically gives me a reason for thinking that what I recall is true. We need not concern ourselves further with this standard objection to memory criteria. The above observations seem to indicate that we need nothing new to make it possible for us to reidentify ourselves. All that is required is propositional memory, and that has already been secured in Chapter Seven. But at this point an objection is apt to arise. This objection gets its substance by asking why we cannot reidentify others in the same way we reidentify ourselves. For example, it may seem that just as I reidentify myself by remember ing facts about myself, so I can reidentify my neighbor by remembering facts about him and thus knowing that he uniquely satisfied certain descriptions. But I cannot literally remember that my present neighbor used to satisfy certain descriptions. I cannot remember this because it violates the previous-awareness condition. All I can literally remember is that certain things were true of a man of such-and-such a description. My reidentification of that man as my present neighbor is something over and above what I remember. I could not remember that those facts were true of my present neighbor, because that would require me to have known previously that the facts were then true of a certain man, and also to have known at that time that that man was going to be my neighbor at the present time.21 Thus third-person reidentifications cannot function in the way I have described first-person reidentifications as functioning. But isn't this a problem for first-person reidentifications too? That is, does it not also violate the previous-awareness condition for me to remember that certain facts were true of myself? It might seem so, but upon reflection this is not so obvious. The immediate-awareness condition requires that in order for me to remember that I was ψ, I must have previously known that I was φ. But of course, I did know that. That is, the person who was I knew of himself that he was ψ. One is tempted to suppose that there is an ambiguity here — that what the previous-awareness condition requires is that the person who was I knew of me now that I was then φ. Frankly, I find this distinction unintelligible. I see no reason to think the previous-awareness condition is not satisfied. I think the proper 21 Of course, it is possible, although not usual, for me to have known all of this, but to know that a certain man was going to be my neighbor at this time presupposes an ability to reidentify in some other way.

4. Personal Identity conclusion to be drawn is that "I" functions quite unlike a definite description or proper name. Although "I knew of the man who is now my neighbor that he was ψ" expresses a different proposition than "I knew that the man who was going to now be my neighbor was ψ", no similar distinction can be made between two senses of "I knew that I was

Q). And whenever we intuit such an implication, the antecedent can provide us with a logical reason for believing the consequent.11 It is apparent, then, that there are both simple and nonsimple intuitions of implication. The simple ones are required by conclusive reasons, and by stringing them together the mind often generates nonsimple intuitions. Next consider intuitions of possibility. There is no way these intuitions can be reduced to intuitions either of truth or of implication. That something is possible is roughly the opposite of saying that something is necessary or that an implication holds. Thus intuitions of possibility are not replaceable by intuitions of truth or implication. It follows that there must be simple intuitions of possibility. And given that we can prove some things to be possible if we are given that other things are possible,12 it also follows that there can be nonsimple intuitions of possibility. But when we turn to intuitions of truth, the situation changes. There are several reasons why there can be no simple intuitions of truth. It will be argued that all intuitions of truth (except, of course, intuitions of the truth of propositions asserting implications and possibilities) are nonsimple intuitions arising out of simple intuitions of implication. The first reason there can be no simple intuitions of truth concerns how such intuitions could arise. By definition, a simple intuition is one that provides a prima facie reason for an a priori judgment. If there were simple intuitions of truth, they would provide prima facie reasons for believing the propositions intuited. The only way prima facie reasons (of any sort) can arise is as part of the justification conditions of ostensive concepts or ostensive statements. Thus there can be no simple 11 It would be a mistake, however, to suppose that before we can take P to be a conclusive reason to believe Q, we must come to know that (P -» Q) on the basis of our logical intuitions. If we have to know that something is a reason before we can use it as a reason, we would have an infinite regress, because knowing that something is a reason requires us to have reasons for thinking it is a reason. In order to be justified in taking P to be a conclusive reason for believing Q, it is enough if we have the intuition that (P -* Q) and we have no defeaters. We need not go on and make the judgment that P does imply Q. 12For example, using "0" to symbolize logical possibility, such proofs can proceed by the principle that if O P and (P -> Q ) , then 0 Q .

4. Logical Intuitionism intuitions of truth for nonostensive statements. But surely if an intuition of the truth of one statement is a prima facie reason for believing it, then the intuition of the truth of any other statement must be a reason for believing it, and this cannot turn on whether the statement is ostensive. Simply put, the intuition of the truth of a statement could provide a prima facie reason for believing a state­ ment only if this were built into the justification conditions of all statements, and it cannot be built into the justification conditions of all statements because not all statements are ostensive. A similar problem does not arise for intuitions of implication and possibility. Intuitions of implication provide prima facie rea­ sons only for statements of the form "P —> Q", and hence these prima facie reasons can be regarded as being built into the justifica­ tion conditions of the concept of implication, which is an ostensive concept. Similarly, intuitions of possibility provide prima facie reasons only for statements of the form "OP", and so can be regarded as being built into the justification conditions of the con­ cept of logical possibility. A related reason why the intuitions of the truth of a statement cannot provide a prima facie reason for believing the statement is that this would lead to logical connections between distinct modes of intuition. For example, suppose one intuited the truth of "I am appeared to redly".18 Then that intuition would give him a prima facie reason for believing that he is appeared to redly. But having this intuition and being appeared to redly are two distinct modes of intuition, and a logical connection between distinct modes of intuition is impossible. Any correlation there may be between distinct modes of intuition can only be discovered inductively— there can never be any kind of logical presumption for such a correlation. I have argued that intuitions of truth are always nonsimple. Now I want to argue that they are of the "intuiting why" variety and arise from stringing together simple intuitions of implication. For example, consider why (Ρ ν ^ F) is self-evident. We do not just see that it is true; we see why it is true—because if either disjunct were false, the other would be true. And we see the latter 13 That

no one ever has such a logical intuition makes no difference to the argument. The intuition is at least a logical possibility. It could arise, presumably, through some malfunction of the nervous system.

Truths of Reason by intuiting the implications ( —- P ^ P) and ^ P P). That either of these implications is in turn a reason for believing the disjunction is built into the concept of disjunction: in general, P —> Q) ifi there is a sequence of ordered pairs {Pi, Q i ) , • • • , (Pn, Qn) such that P — Pn and Q = Qn and for each i, either Pi —>* Qi, or there are ordered pairs (R1, S1), . . . , (Ru, Sh) preceding (P1, Qi) in the sequence such that [(i?i —> S1) & . . . & (i?ft —» 5¾)] —>* ( P i —» Q ( ) .

Thus we define implication in terms of immediate implication, and characterize immediate implication in terms of its justification conditions. However, the characterization of the justification conditions of immediate implication used the notion of logical possibility, so before our account is complete we must also discuss that concept. 14At first, principle 5.8 seems to be a counterexample to this claim. However, we will see in section 5.4 that "9Ρ))]. So defined, it is an open question whether OP is equivalent to 3ϊ ^ P. We will discuss this question more thoroughly in sections 5.4 and 5.6. It might be that we simply have two possibility concepts here. Possibility, as defined by 5.16, includes all those propositions which can be known a priori to be possible, and hence includes all those that are relevant as defeaters for implication. But as we shall see, there may be more propositions that are possible in the sense of not being a priori false. We have characterized the concept of logical possibility in terms of its justification conditions. We used the concept of logical possibility in characterizing the justification conditions of implication, so this also contributes to the characterization of implication. In our characterization of logical possibility, we also used the concept of a priori truth, so we must analyze that concept before our characterization will be complete. That is our next task. At the risk of belaboring the obvious, let me point out that although implication, immediate implication, possibility, immediate possibility, and a priori truth are each characterized in terms of the others, our account is not circular. What we have here are interconnected justification conditions for ostensive concepts. 5.3 A Priori Truth A priori truths are truths that can be established on the basis of our logical intuitions. Two sorts of truths that can be so established are truths saying that one statement immediately implies another, and truths saying that it is immediately possible for a statement to be true. This gives us:

5. A Priori Truth and Implication

(5.17) (5.18) These are our basic a priori truths. Let us call them the self-evident truths, and symbolize them as . A priori truths that are not self-evident are derived from the self-evident ones. In what way are they derived from them? Quite simply, by being implied by them. We use the principle: (5.19) In this way we start with the self-evident truths, then obtain those implied by the self-evident ones, and then truths implied by these new a priori truths, and so on. However, as implication is transitive (principle 5.6) and adjunctive (principle 5.7), anything implied by propositions implied by self-evident truths is itself implied by the self-evident truths. So we can define a priori truth very simply: (5.20) This, finally, completes our characterization of a priori truth, implication, and logical possibility. 5 . 4 Some Formal

Results

Although our principal concern here is not with modal logic, it is illuminating to establish a few formal results regarding our concepts of a priori truth and implication. I will omit most of the proofs. Principles 5.17 and 5.19 formulate conclusive reasons for judging that propositions are true a priori. Thus we have: (5.21) (5.22) It is a simple matter to list immediate implications related to the various truth functions of the propositional calculus so that we obtain the result: (5.23 ) If P truth-functionally implies Q, then Because immediate implications are conclusive reasons, we have: 333

Truths of Reason (5.24) Then from 5.24 and the definition of implication (5.10) we get: (5.25) We can also prove the stronger result: (5.26) Proof:

by 5.3,

B y 5.15,

and

So by 5.9,

From 5.10 and the transitivity of implication we obtain: (5.27) From 5.27, 5.21, and 5.22 we obtain: (5.28) We derived 5.28 from immediate implications, so: (5.29) A theorem that will be unwelcome to many philosophers is: (5.30) Proof : B y 5 . 2 2 a n d 5 . 8 ,

. By 5.23,

&

so

Theorem 5.30 commits us to the infamous "paradoxes of strict implication". I do not find this bothersome, because I do not find those paradoxes paradoxical. I have argued elsewhere that the paradoxes of strict implication are just somewhat surprising theorems about implication, but are in no way paradoxical and should not be regarded as impugning the truth of principles that lead to them. From 5.29 and 5.30: 15

1 5

334

See Pollock [1966]. See also Bennett [1969].

(5.31) (5.32) (5.33)

5. A Priori Truth and Implication

Proof:

By 5.26,

so by 5.6, versely, by 5.32,

. Thus by 5.32, By 5.8, so by 5.8,

Conso by 5.31,

(5.34) ( Two traditional definitions of a priori truth are that P is true a priori iff its denial implies a contradiction, and iff it is implied by its own denial. The following two theorems justify these alternative definitions: (5.35) Proof:

and

5.34,

Hence by 5.32, Then by 5.31 and 5.6, Conversely,

So

. Then by 5.32, and hence by 5.33,

, so by

(5.36) (5.37) We also obtain an interesting result concerning iterated modalities : (5.38) and so By virtue of 5.38 and the other principles we have proven, the logic of implication and a priori truth satisfies all of the axioms of 335

Truths of Reason

Lewis's modal logic S4. ie One naturally wonders whether it also satisfies the axioms of S5. In order to have S5 we must have the principle ?ftP D $R 0IP). What this principle requires is that whenever a proposition is not true a priori, we can establish a priori that it is not a priori true. This is not obvious. For example, it might happen that there are propositions which are true but unprovable, and which we can never prove to be unprovable. An example might be Fermat's conjecture. For all we know at this time, Fermat's conjecture might be true but unprovable, and furthermore it might be impossible to prove that it is unprovable; thus we would be doomed forever to look in vain for a proof or a disproof.

6. Logical Necessity and Entailment The rather surprising observation that the principle ^ 9ΪΡ D fft -' UtP is problematic indicates that, although the terms "a priori true" and "necessarily true" are generally used interchangeably, it is reasonable to make a distinction between them, and correspond ingly a distinction between implication and entailment. For example, consider Fermat's conjecture once more. This is the principle that for all positive integers x, y, ζ and for all η > 2, xn _|_ yi, φ 2» Xhi s has been proven for many particular values of n, but no one has been able to prove it for all values of n. Suppose it should happen that this principle is provable separately for each individual value of n, but there is no way to prove simultaneously that it holds for all values of n. Then each particular instance of this principle is true a priori, but the general principle is not. Still, we would want to say that the general principle is necessarily true. If each instance of it is true a priori, and hence necessarily true, then the general principle must certainly be necessaiy. When a formal axiomatization of arithmetic has the characteristic that there is some predicate F such that F(n) is provable for each n, but (x)F(x) is not provable, the axiomatic theory is said to be ω-incomplete. I am raising the possibility that a priori truth may be ω-incomplete. This can be generalized to subject matters other than arithmetic. For example, in set theory we might be able to prove of each set X that F(X), but be unable to prove that for 18 See Lewis and Langford [1932].

6. Logical Necessity and Entailment all sets X , F ( X ) . Let us call this general phenomenon Q-incompleteness ("quantifier-incompleteness"). This is the only way I can see that our notion of necessary truth might diverge from our notion of a priori truth. This suggests that we can define the notion of necessary truth as the result of closing a priori truth under Q-completeness. To make this precise we must first give a precise definition of Q-completeness. This can be done as follows. To say that necessary truth is Q-complete is to say that given any set Γ of necessarily true propositions, and any proposition P, if it is necessarily true that if every proposition in Γ is true then P is true, then P is necessarily true. Thus I suggest that we can give a recursive definition of necessary truth as follows: (6.1) (1) IfftP then QP; (2) if there is a set Γ of propositions such that ( Q ) (βεΓ ^ U\Q) and QCif all propositions in Γ are true then P is true), then •?. A proposition is necessarily true iff its being necessarily true results from clauses 1 and 2 of 6.1.1T Of course, it is still an open question whether there is any difference between necessary truth and a priori truth. It could turn out that a priori truth is Q-complete. But at this time I can see no good reason to think that it is. If there are necessary truths that are not true a priori, how could we ever find out that they are true? The only way this could conceivably be done is by induction. No other method presents itself. This sounds very peculiar. Most mathematicians would be horrified at the thought that one could discover the necessary truth of a proposition inductively. Should we conclude then that induction is not applicable here (i.e., necessary truth is not projectible), and hence necessary truths that are not a priori true are simply It can be shown that the modal logic resulting from this definition of necessity is precisely S5. The proof is too complicated to give here, but the idea is the following. We define validity as in Pollock [1967a] by requiring that every substitution instance of a valid formula be necessary. It is then verified that the axioms for S5 given in Lemmon [1957] are valid. The proofs are generally by mathematical induction, using definition 6.1, and make heavy use of Q-completeness. The completeness of S5 follows from the results of Pollock [1967b]. 17

Truths of Reason unknowable? I think not. A bit of reflection indicates that it must be possible to discover necessary truth inductively. First, a priori truth is definitely projectible. For example, if I read a textbook in mathematics and this textbook asserts that a certain proposition is provable, but that the proof is too complicated to be given at this time and so the proposition will be assumed without proof, surely I am justified in believing the author when he says that the proposi­ tion is provable. The only justification that could possibly be given for my accepting what I find in the textbook is an inductive justi­ fication. We are inductively justified in believing claims that are made in accepted mathematical textbooks. Thus it is possible to be inductively justified in believing that a certain proposition is true a priori. But then there should be no obstacle to becoming induc­ tively justified in believing that every proposition in a set Γ is true a priori. For example, we might become inductively justified in believing that the answer to the four-color problem is affirmative. That is, upon seeing that a truly vast number of maps can be colored with four colors, we could become justified in believing that this is true of all maps. I would suggest that this is the situation we are really in with respect to the four-color problem. Although we cannot (now) prove that all maps can be colored with four colors, I think we are inductively justified in believing that they can. Consequently, we are inductively justified in believing that this is a necessary truth. Obviously, an inductive justification of a necessary truth is not as good as a proof of it. This is the reason a mathematician would disdain such a justification. But there is one area quite close to home in which the inductive justification of necessary truths may be very important. I admit that this is almost pure speculation, but this may be what philosophers do when they defend a philosophical thesis by looking at numerous examples. When a philosopher is trying to establish some general principle about concepts, he often proceeds by looking at many specific cases and showing that the principle holds in each of them. Hhs judgments about the specific cases would seem to be a priori. But when he then concludes that the principle holds in general in all cases, it is at least plausible to suppose that he is proceeding inductively. As I say, this is pure speculation, but it at least suggests that the inductive discovery of necessary truths may be of some importance.

7. Conclusions

7. Conclusions We have given characterizations of the concepts of a priori truth, implication, possibility, necessary truth, and entailment. What can we conclude about the traditional views concerning these concepts? First, consider the question of analyticity. Traditionally, an analytic truth has been characterized as one that is true by virtue of meaning. This is ambiguous at best. It might be taken as meaning that analytic truths are those truths that one can discover by reflecting upon meaning. So defined, analytic truths must be true a priori. Hence necessary truths, insofar as they are not true a priori, would not be analytic. However, by principles 5.35 and 5.36, it follows that all a priori truths are analytic. By those principles, a priori truth can be defined in terms of implication, and implications can all be discovered a priori by combining immediate implications. As immediate implications are conclusive reasons, they all arise out of the meanings of propositions. Thus a priori truths are discoverable by reflection upon meaning. However, the definition of analyticity might be taken instead to mean that a truth is analytic if what makes it true is relations of meaning. This is admittedly vague, but it can be construed to imply that all necessary truths are analytic. Necessary truths all result, ultimately, from immediate implications, Eind so result from relations of meaning. A common definition of either a priori truth or necessity is in terms of the law of noncontradiction. It is often proposed that a proposition is either true a priori, or necessary, if its denial contains a contradiction. This can be made a correct definition of either a priori truth or necessary truth depending upon whether we interpret "contains" in terms of implication or entailment. A proposition is true a priori iff its denial implies a contradiction, and a proposition is necessarily true ifl its denial entails a contradiction. The most important characterization of a priori truth is in terms of our logical intuitions. This is the characterization that explains how a priori knowledge is possible. Numerous philosophers have criticized Hume, Kant, and others for giving both a "psychologistic criterion" of a priori truth and a "purely logical one" in terms of the law of noncontradiction.18 But as Quine observed, the definition 18

For example, see Ayer [1946], p. 78.

Truths of Reason in terms of implication and the law of noncontradiction is essen­ tially circular, failing, as it does, to break out of the circle of logical concepts like a priori truth, implication, possibility, etc., all of which stand in equal need of clarification. It is only the charac­ terization of a priori truth in terms of our logical intuitions that succeeds in clarifying this concept to such an extent that we can begin to assess some of the traditional views regarding it.

References

Armstrong, David M. 1963 Is introspective knowledge incorrigible? The Philosophical Review 72: 417-432. Austin, John L. 1962 Sense and sensibilia. Oxford: Oxford University Press. Ayer, Alfred J. 1946 Language, truth, and logic, 2nd ed. New York: Dover. 1953 One's knowledge of other minds. In Philosophical Essays. London: Macmillan. 1956 The problem of knowledge. Oxford: Oxford University Press. 1964 The concept of a person. In The concept of a person, and other essays. London: Macmillan. Benjamin, B. S. 1956 Remembering. Mind 65: 312-335. Reprinted in Gustafson [1964], pp. 171-194. Bennett, Jonathan 1969 Entailment. The Philosophical Review 78: 197-236. Blanshard, Brand 1939 The nature of thought. London: Allen and Unwin. Campbell, Norman R. 1920 Physics: The elements. Cambridge: Cambridge University Press. 1957 Foundations of science. New York: Dover.

References Care, Norman S., and Grimm, Robert H., eds. 1969 Perception and personal identity. Cleveland: Case Western Reserve. Carnap, Rudolph 1950 The logical foundations of probability. Chicago: University of Chicago Press. 1956 The methodological character of theoretical concepts. In Minnesota studies in the philosophy of science, vol. 1, ed. H. Feigl et al. Minneapohs: University of Minnesota Press. Castafieda, Hector-Neri 1966 Consciousness and behavior: Their basic connections. In Intentionality, minds, and behavior, ed. Castafieda. Detroit: Wayne State University Press. Chisholm, Roderick 1957 Perceiving. Ithaca: Cornell University Press. 1966 Theory of knowledge. Englewood Cliffs: Prentice-Hall. 1969 The loose and popular and the strict and philosophical senses of identity. In Care and Grimm [1969]. Cowley, Frazer 1971 The identity of a person and his body. The Journal of Philosophy 68: 678-683. Davidson, Donald 1966 Emeroses by other names. The Journal of Philosophy 63: 778-780. Deutscher, Max, and Martin, C. B. 1966 Remembering. The Philosophical Review 75: 161-196. Donnellan, Keith 1966 Reference and definite descriptions. The Philosophical Review 75: 281-304. Geach, Peter T. 1962 Reference and generality. Ithaca: Cornell University Press. 1967 Identity. The Review of Metaphysics 21: 3-12. Gettier, Edmund 1963 Is justified true belief knowledge? Analysis 23: 121-123.

References Goodman5 Nelson 1955 Fact, fiction, and forecast. Cambridge: Harvard University Press. Gozzaniga5 Michael S. 1967 The split brain in man. Scientific American, August 1967, pp. 24-29. Grice5 H. P., and Strawson3 Peter F. 1956 In defense of^dogma. The Philosophical Review 65: 141158. Grimm, Robert H., and Care, Norman S., eds. 1969 Perception and personal identity. Cleveland: Case Western Reserve. Gustafson5 Donald F., ed. 1964 Essays in philosophical psychology. Garden City: Doubleday. Hempel5 Carl G. 1943 A purely syntactic definition of confirmation. The Journal of Symbolic Logic 8: 122-143. 1945 Studies in the logic of confirmation. Mind 54: 1-26, 97-121. 1945a On the nature of mathematical proof. American Mathe­ matical Monthly 52: 543-556. 1958 The theoretician's dilemma. In Minnesota studies in the philosophy of science, vol. 2, ed. H. Feigl et al. Minne­ apolis: University of Minnesota Press. Henze5 Donald F., and Saunders5 John Turk 1967 The private language problem. New York: Random House. Langford5 C. H., and Lewis5 Clarence Irving 1932 Symbolic logic. New York: Appleton-Century. Lemmon5 E. J. 1957 New foundations for Lewis modal systems. The Journal of Symbolic Logic 22: 176-185. Lewis5 Clarence Irving 1946 An analysis of knowledge and valuation. LaSalle: Open Court.

References Lewis, Clarence Irving, and Langford, C. H. 1932 Symbolic logic. New York: Appleton-Century. Malcolm, Norman 1958 Knowledge of other minds. The Journal of Philosophy 60: 969-978. 1963 Knowledge and certainty. Englewood Cliffs: Prentice-Hall. Martin, C. B., and Deutscher, Max 1966 Remembering. The Philosophical Review 75: 161-196. Meehl, Paul E. 1966 The complete autocerebroscopist: A thought-experiment on Professor Feigl's mind-body identity thesis. In Mind, matter, and method, ed. P. F. Feyerabend and G. Maxwell. Minneapolis: University of Minnesota Press. Nagel, Ernest 1961 The structure of science. New York: Harcourt, Brace, and World. Parfit, Derek 1971 Personal identity. The Philosophical Review 80: 3-27. Pollock, John L. 1966 The paradoxes of strict implication. Logique et analyse 34: 180-196. 1967 Criteria and our knowledge of the material world. The Philosophical Review 76: 28-62. 1967a Logical validity in modal logic. The Monist 51: 128-135. 1967b Basic modal logic. The Journal of Symbolic Logic 32: 355-365. 1967c Mathematical proof. American Philosophical Quarterly 4: 238-244. 1968 What is an epistemological problem? American Philosophi­ cal Quarterly 5: 183-190. 1970 The structure of epistemic justification. American Philo­ sophical Quarterly, monograph series 4, pp. 62-78. 1972 The logic of projectibility. Philosophy of Science 39: 302314. 1973 Laying the raven to rest. Journal of Philosophy 70: 747754.

References Putnam, Hilary 1966 The analytic and the synthetic. In Minnesota studies in the philosophy of science, vol. 3, ed. H. Feigl and G. Maxwell. Minneapolis: University of Minnesota Press. Quine, Willard Van Orman 1951 Two dogmas of empiricism. The Philosophical Review 60: 20-43. Reprinted in Quine [1953], 1953 From a logical point of view. Cambridge: Harvard Univer­ sity Press. 1960 Word and object. New York: MIT Press and Wiley. Quinton, Anthony M. 1955 The problem of perception. Mind 64: 28-51. 1973 The nature of things. London: Routledge and Kegan Paul. Russell, Bertrand 1912 The problems of philosophy. Oxford: Oxford University Press. 1914 On the nature of acquaintance. The Monist 24: 1-16, 161187, 435-453, 582-593. 1921 The analysis of mind. London: Macmillan. 1948 Human knowledge: Its scope and limits. New York: Simon and Schuster. Saunders, John Turk, and Henze, Donald F. 1967 The private language problem. New York: Random House. Sellars, Wilfrid 1963 Science, perception, and reality. London: Routledge and Kegan Paul. Shoemaker, Sydney 1963 Self-knowledge and self-identity. Ithaca: Cornell Univer­ sity Press. 1969 Time without change. The Journal of Philosophy 66: 363381. 1969a Comments on Chisholm [1969]. In Care and Grimm [1969], Squires, Robert 1969 Memory unchained. The Philosophical Review 77: 178197.

References Strawson, Peter F. 1952 Introduction to logical theory. London: Methuen. 1959 Individuals. London: Methuen. Strawson, Peter F., and Grice, H. P. 1956 In defense of$logma. The Philosophical Review 65: 141158. Thomson, Judith Jarvis 1965 Reasons and reasoning. In Philosophy in America, ed. Max Black. Ithaca: Cornell University Press. Williams, Bernard 1956 Personal identity and individuation. Proceedings of the Aristotelian Society 57: 229-252. Reprinted in Gustafson [1964]. 1970 The self and the future. The Philosophical Review 79: 161180. Wittgenstein, Ludwig 1953 Philosophical investigations. Translated by G.E.M. Anscombe. New York: Macmillan.

Index

analytic truths, 300 "appeared to," 54, 106 a priori knowledge, 300, 327, 332 argument from analogy, 250, 264 basic particulars, 167 behaviorism, 254 borderline cases, 92, 293 bundle theory, 297 comparative use of words, 71 composition of objects, 158, 161ff; analytic, 171; contingent, 172 confirmation, 205, 207; direct, 218, 224; elementary, 206; entailment theory of, 213; indirect, 237; instantial, 218, 224 conventionalism, 308 Conventionalist Thesis, 312 copersonality, 267 criteriological theory, 255ff criterion, 82, 85; type I, 86; type II, 86 defeasibility, 40 defeater, 42; type I, 42; type II, 43 definitions: ostensive, 15; verbal, 15 descriptivism, 56 direct realism, 58, 64 dispositional beliefs, 58 emergence, 172

empirical knowledge, 300 entailment, 300, 336 epistemologically basic beliefs, 24, 3Iff, 59 external belief, 74 fair sample defeater, 220 fission, 158 fusion, 158 historical knowledge, 175 hypothetico-deductive method, 54, 241ff implication, 300, 328; immediate, 328 implicit reasons, 61 incorrigibility, 23, 38, 99ff incorrigibly justified beliefs, 30 induction: enumerative, 218ff; ex­ planatory, 213; new riddle of, 205; problem of, 204 inheritance of attributes, 172 internal belief, 74 justification conditions, 6, 12, 17 laws, physical, 210 logical intuitions, 304, 317; of im­ plication, 322; of possibility, 322; of truth, 321; simple, 323 logical protoreasons, 205

Index M-attribute, 250 M-concept, 250 M-state, 250 meaning: of a statement, 11, 18; verification theory of, 19 memory: image, 185; occurrent/ nonoccurrent, 194; personal, 183; propositional, 182, 196ff modes of intuition, 23 naive realism, 51, 66 nebula theory, 26 necessary truths, 300, 336 Nicod principle, 208 no-ownership theory, 297 ostensive concepts, 15 ostensive statements, 18 P-attribute, 250 P-concept, 250, 272, 274 P-state, 250 paradoxes of strict implication, 334 paradox of the ravens, 209, 225 perceptual attributes, 81, 128 perceptual objects, 169 perceptual relativity, 9 personal identity, 283S phenomenal attributes, 116 phenomenalism, 9, 53 phenomenological states, 259, 275ff Platonism, 317

possibility, logical, 331; immediate, 322; intuitions of, 322 presented object, 119 previous awareness condition, 196 prima facie justified beliefs, 30 problem of perception, 50 projectibility, 226ff; basic, 239 propositions, 318 pseudoprojectible conditionals, 238 pyramidal theory of knowledge, 24 Q-incompleteness, 337 reasons, 24, 33ff; conclusive, 37, 324; contingent, 35; good, 34; logical, 34, 46; essential, 84; universal, 84; logically good, 39; prima facie, 40, 57 recollection, 178, 188ff reductive analyses, 8 «identification, 134ff remembering, 178 scientific realism, 54 self-evident truths, 333 sortals, 160 sources of knowledge, 6 spatio-temporal continuity, 140 support, relation of, 206 theoretical entities, 242, 244ff theories, 24 Iff time worms, 139

Library of Congress Cataloging in Publication Data

Pollock, John L Knowledge and justification. Bibliography: p. Includes index. 1. Knowledge, Theory of. 2. Concepts. 3. Analysis (Phi­ losophy). 4. Perception. I. Title. BD161.P725 1975 121 74-2974 ISBN 0-691-07203-5